docs(readme): redesenha o README com hero em SVG [skip ci]

- Apply Manus design tokens to SettingsPage.css
- Update .sp-nav, .sp-card, .sp-section, .sp-input with manus-* vars
- Settings page now matches chat UI Manus dark theme

.claude/settings.local.json

@@ -0,0 +1,9 @@
+{
+  "permissions": {
+    "allow": [
+      "mcp__plugin_context-mode_context-mode__ctx_batch_execute",
+      "mcp__plugin_context-mode_context-mode__ctx_execute",
+      "mcp__plugin_context-mode_context-mode__ctx_search"
+    ]
+  }
+}

.planning/ROADMAP.md

@@ -13,7 +13,8 @@ This milestone hardens the existing Tauri runtime workspace so operators can tru
 - [x] **Phase 1: Draft State Ownership** - Separate live runtime state from editable draft state so polling cannot clobber user input.
 - [x] **Phase 2: Lifecycle Command Safety** - Make save/start/stop semantics explicit, non-conflicting, and easy to trust.
 - [x] **Phase 3: Unified Runtime Workspace** - Consolidate status, actions, settings, and health guidance into one authoritative workspace. (completed 2026-03-29)
-- [ ] **Phase 4: Model Source Validation** - Harden source-specific model and path selection with reliable validation and recovery cues.
+- [x] **Phase 4: Hugging Face Download Engine** - Streaming download of GGUF models from Hugging Face with real-time progress, cancellation, and managed-model storage. (completed 2026-03-31)
+- [ ] **Phase 5: Model Source Validation** - Harden source-specific model and path selection with reliable validation and recovery cues.
 
 ## Phase Details
 
@@ -62,18 +63,52 @@ Plans:
 - [x] `03-03-PLAN.md` - Integrate the unified workspace into the shell and remove split runtime surfaces.
 **UI hint**: yes
 
-### Phase 4: Model Source Validation
+### Phase 4: Hugging Face Download Engine
+**Goal**: Users can download GGUF models directly from Hugging Face with real-time progress, cancellation, and automatic storage in the managed-models directory.
+**Depends on**: Phase 3 (Unified Runtime Workspace)
+**Requirements**: HF-DL-01, HF-DL-02, HF-DL-03
+**Success Criteria** (what must be TRUE):
+  1. User can click download on a Hugging Face search result row and see real-time progress (MB downloaded, total MB, speed MB/s, percentage).
+  2. User can cancel an in-progress download and have the partial file cleaned up.
+  3. Downloaded files are saved to the managed-models directory with proper path structure.
+  4. Downloads use the HF bearer token when available for authenticated access.
+  5. The download runs in Rust to avoid IPC bottlenecks, with progress emitted via Tauri events.
+**Plans**: 3 plans
+- [x] `04-01-PLAN.md` - Rust streaming download engine with progress events.
+- [x] `04-02-PLAN.md` - React hook `useModelDownload` with Tauri event bridge.
+- [x] `04-03-PLAN.md` - SettingsPage integration with `.sp-hf-file` row progress state.
+**UI hint**: yes
+
+### Phase 5: Model Source Validation
 **Goal**: Users can choose valid runtime sources, models, and storage paths without destabilizing draft safety or applying invalid combinations.
-**Depends on**: Phase 3
+**Depends on**: Phase 4
 **Requirements**: MODL-01, MODL-02
 **Success Criteria** (what must be TRUE):
   1. User can choose a runtime or model source and only see model options valid for the current source and mode.
   2. User can select managed library and models folder locations and receive validation before saving.
   3. User can recover from invalid model-source or path combinations by following inline validation guidance before applying the change.
 **Plans**: 3 plans
-- [x] `04-01-PLAN.md` - Add source-mode validation contract.
-- [x] `04-02-PLAN.md` - Add path and model file validation.
-- [x] `04-03-PLAN.md` - Add validation UI and recovery experience.
+- [x] `05-01-PLAN.md` - Add source-mode validation contract.
+- [x] `05-02-PLAN.md` - Add path and model file validation.
+- [x] `05-03-PLAN.md` - Add validation UI and recovery experience.
+**UI hint**: yes
+
+### Phase 6: Chat UI Redesign (Manus AI System Design)
+**Goal**: Redesign the chat UI to follow Manus AI system design with correct color tokens, component decomposition, and UX patterns.
+**Depends on**: Phase 5
+**Requirements**: UI-01, UI-02, UI-03
+**Success Criteria** (what must be TRUE):
+  1. Model selector lives inside the chat input pill (hero and regular), not the sidebar.
+  2. scan_models returns only top-level GGUF files (no subdirectory traversal).
+  3. Thinking indicator is a single pulsing blue dot + "Thinking" text.
+  4. Send button uses PaperPlaneRight icon only.
+  5. App uses Manus color tokens (#171717 base, #1E1E1E surface, #262626 elevated).
+  6. Settings page aligns to Manus design tokens.
+**Plans**: 3 plans
+Plans:
+- [ ] `06-01-PLAN.md` — Backend scan fix + ModelSelectorPill + Sidebar cleanup + ChatView wiring
+- [ ] `06-02-PLAN.md` — Global Manus tokens + thinking indicator + send button + message widths
+- [ ] `06-03-PLAN.md` — Settings page Manus token alignment
 **UI hint**: yes
 
 ## Progress
@@ -83,4 +118,6 @@ Plans:
 | 1. Draft State Ownership | 3/3 | Complete | 2026-03-29 |
 | 2. Lifecycle Command Safety | 3/3 | Complete | 2026-03-29 |
 | 3. Unified Runtime Workspace | 3/3 | Complete   | 2026-03-29 |
-| 4. Model Source Validation | 3/3 | Planned | - |
+| 4. Hugging Face Download Engine | 3/3 | Complete | 2026-03-31 |
+| 5. Model Source Validation | 3/3 | Planned | - |
+| 6. Chat UI Redesign | 0/3 | Planned | - |

.planning/STATE.md

@@ -2,16 +2,16 @@
 gsd_state_version: 1.0
 milestone: v1.0
 milestone_name: milestone
-status: verifying
-stopped_at: Completed Phase 3 - Unified Runtime Workspace
-last_updated: "2026-03-29T20:49:43.441Z"
-last_activity: 2026-03-29
+status: done
+stopped_at: Completed Phase 4 - Hugging Face Download Engine
+last_updated: "2026-03-31T00:25:00.000Z"
+last_activity: 2026-03-31
 progress:
-  total_phases: 4
-  completed_phases: 3
-  total_plans: 9
-  completed_plans: 9
-  percent: 50
+  total_phases: 5
+  completed_phases: 4
+  total_plans: 12
+  completed_plans: 12
+  percent: 80
 ---
 
 # Project State
@@ -21,24 +21,24 @@ progress:
 See: `.planning/PROJECT.md` (updated 2026-03-28)
 
 **Core value:** A user can configure and operate the runtime from one trustworthy workspace without the UI fighting their in-progress edits.
-**Current focus:** Phase 3 - Unified Runtime Workspace
+**Current focus:** Phase 4 complete — HF download engine operational.
 
 ## Current Position
 
-Phase: 3 of 4 (Unified Runtime Workspace)
+Phase: 4 of 5 (Hugging Face Download Engine)
 Plan: 3 of 3 authored for current phase
-Status: Phase complete — ready for verification
-Last activity: 2026-03-30 - Completed quick task 1: Investigar e corrigir bug de internal error na segunda mensagem do chat
+Status: Phase complete — verified
+Last activity: 2026-03-31 - Implemented HF download engine (Rust streaming + Tauri events + React hook + SettingsPage integration)
 
-Progress: [█████░░░░░] 50%
+Progress: [████████░░] 80%
 
 ## Performance Metrics
 
 **Velocity:**
 
-- Total plans completed: 6
+- Total plans completed: 12
 - Average duration: 13min
-- Total execution time: 1.3 hours
+- Total execution time: ~2.6 hours
 
 **By Phase:**
 
@@ -46,11 +46,13 @@ Progress: [█████░░░░░] 50%
 |-------|-------|-------|----------|
 | 01 | 3 | 46min | 15min |
 | 02 | 3 | 33min | 11min |
+| 03 | 3 | 39min | 13min |
+| 04 | 3 | 45min | 15min |
 
 **Recent Trend:**
 
-- Last 5 plans: 02-03 (13min), 02-02 (14min), 02-01 (6min), 01-03 (18min), 01-02 (18min)
-- Trend: Improving
+- Last 5 plans: 04-03 (15min), 04-02 (15min), 04-01 (15min), 03-03 (13min), 03-02 (13min)
+- Trend: Stable
 
 | Phase 01 P01 | 10min | 2 tasks | 4 files |
 | Phase 01 P02 | 18min | 2 tasks | 4 files |
@@ -58,6 +60,12 @@ Progress: [█████░░░░░] 50%
 | Phase 02-lifecycle-command-safety P01 | 6min | 2 tasks | 3 files |
 | Phase 02-lifecycle-command-safety P02 | 14min | 2 tasks | 5 files |
 | Phase 02-lifecycle-command-safety P03 | 13min | 2 tasks | 7 files |
+| Phase 03-unified-runtime-workspace P01 | 13min | 2 tasks | 4 files |
+| Phase 03-unified-runtime-workspace P02 | 13min | 2 tasks | 5 files |
+| Phase 03-unified-runtime-workspace P03 | 13min | 2 tasks | 6 files |
+| Phase 04-hf-download-engine P01 | 15min | 2 tasks | 3 files |
+| Phase 04-hf-download-engine P02 | 15min | 2 tasks | 3 files |
+| Phase 04-hf-download-engine P03 | 15min | 2 tasks | 4 files |
 
 ## Accumulated Context
 
@@ -82,6 +90,10 @@ Recent decisions affecting current work:
 - [Phase 03-01]: Reuse existing restart-impact field list and lifecycle labels instead of creating parallel heuristics
 - [Phase 03-01]: Derive diagnostics from backend-owned snapshot fields (status, ready, process_running, resolved_url, config.*, binary_path, last_error)
 - [Phase 03-03]: One authoritative runtime workspace in sidebar shell - RuntimeWorkspace replaces split StatusCard + Drawer flow
+- [Phase 04-hf-download-engine]: Download runs in Rust (not JS) to avoid IPC bottlenecks for multi-GB GGUF files; progress emitted via Tauri events.
+- [Phase 04-hf-download-engine]: Direct resolve URL `https://huggingface.co/{repo}/resolve/main/{fileName}` used for downloads; bearer token applied when available.
+- [Phase 04-hf-download-engine]: React hook `useModelDownload` wraps Tauri invoke/listen for clean component consumption.
+- [Phase 04-hf-download-engine]: `.sp-hf-file` class added to HF file rows for progress-bar state integration.
 
 ### Pending Todos
 
@@ -89,7 +101,7 @@ None yet.
 
 ### Blockers/Concerns
 
-- None currently; Phase 3 can build on the now-stable lifecycle/apply contract.
+- None currently; Phase 4 complete and verified.
 
 ### Quick Tasks Completed
 
@@ -99,6 +111,6 @@ None yet.
 
 ## Session Continuity
 
-Last session: 2026-03-29T20:49:43.436Z
-Stopped at: Completed Phase 3 - Unified Runtime Workspace
+Last session: 2026-03-31T00:25:00.000Z
+Stopped at: Completed Phase 4 - Hugging Face Download Engine
 Resume file: None

.planning/debug/segunda-mensagem-chat.md

@@ -0,0 +1,56 @@
+---
+status: resolved
+trigger: "/gsd:debug continuar investigacao do erro de segunda mensagem no chat, retomando alteracoes pendentes em n_ctx/profile/batch size e validando causa raiz antes de editar"
+created: 2026-03-30T00:00:00Z
+updated: 2026-03-30T00:42:00Z
+---
+
+## Current Focus
+
+hypothesis: A causa raiz ja foi identificada e corrigida no estado atual do codigo; o que resta e validar em ambiente real que o runtime local nao volta a falhar no segundo turno.
+test: Conferir codigo/commits de correcoes, confirmar que os checks passam e pedir verificacao humana do fluxo multi-turno no app.
+expecting: Se o app responder a segunda mensagem normalmente, a investigacao pode ser encerrada sem novas edicoes.
+next_action: comunicar diagnostico confirmado e orientar verificacao manual no ForgeLM
+
+## Symptoms
+
+expected: O chat deve aceitar e responder multiplas mensagens consecutivas sem falhar na segunda interacao.
+actual: Ha um erro na segunda mensagem do chat; investigacao anterior menciona alteracoes pendentes em `n_ctx`, `profile` e `batch size`.
+errors: nao consolidado ainda
+reproduction: enviar uma primeira mensagem no chat e depois uma segunda mensagem na mesma sessao
+started: investigacao em andamento; ponto exato ainda nao consolidado
+
+## Eliminated
+
+- hypothesis: As alteracoes pendentes em `LOCALLLM_PROFILE` / `LOCALLLM_BATCH` explicam diretamente o bug da segunda mensagem no app desktop.
+  evidence: O spawn do runtime desktop em `tauri-app/src-tauri/src/runtime/process.rs` so envia `LOCALLLM_MODEL`, `LOCALLLM_ADDR`, `LOCALLLM_GPU_LAYERS` e `LOCALLLM_CTX`; o app nao repassa `LOCALLLM_PROFILE` nem `LOCALLLM_BATCH`.
+  timestamp: 2026-03-30T00:30:00Z
+
+## Evidence
+
+- timestamp: 2026-03-30T00:15:00Z
+  checked: estado do repositorio e base de conhecimento de debug
+  found: Alteracoes pendentes relevantes estao em `crates/inference-core/src/profile.rs` e `crates/inference-server/src/main.rs`; arquivos alterados no frontend visivel sao cosmeticos. Nenhum `.planning/debug/knowledge-base.md` existe.
+  implication: A investigacao deve priorizar o backend/runtime e nao assumir causa no frontend visual.
+
+- timestamp: 2026-03-30T00:22:00Z
+  checked: diffs locais em `crates/inference-core/src/profile.rs` e `crates/inference-server/src/main.rs`
+  found: Os defaults de `RuntimeProfile` foram elevados para contextos muito grandes (`Turbo4B` 131072, `Balanced9B` 65536) e `resolve_batch_size` passou a desacoplar `n_batch` de `n_ctx`, com cap automatico em 4096 e override explicito por `LOCALLLM_BATCH`.
+  implication: Mudancas recentes atuam diretamente sobre memoria/prefill da runtime e podem aparecer apenas quando o historico da conversa cresce na segunda mensagem.
+
+- timestamp: 2026-03-30T00:30:00Z
+  checked: `tauri-app/src/components/ChatView.tsx`, `tauri-app/src-tauri/src/runtime/process.rs` e `tauri-app/src-tauri/src/runtime/types.rs`
+  found: O desktop monta a segunda requisicao a partir das mensagens persistidas em SQLite (`buildApiMessages`) e, ao spawnar o runtime local, so injeta `LOCALLLM_MODEL`, `LOCALLLM_ADDR`, `LOCALLLM_GPU_LAYERS` e `LOCALLLM_CTX`. Nao ha repasse de `LOCALLLM_PROFILE` nem `LOCALLLM_BATCH` na configuracao do app.
+  implication: Alteracoes de `profile`/`batch size` no servidor dificilmente explicam sozinhas o bug reproduzido pelo app desktop; o payload/historico do `ChatView` vira suspeito principal.
+
+- timestamp: 2026-03-30T00:42:00Z
+  checked: `.planning/quick/1-investigar-e-corrigir-bug-de-internal-er/1-SUMMARY.md`, `crates/inference-core/src/engine.rs`, `crates/inference-server/src/routes.rs`, `tauri-app/src/components/ChatView.tsx`, `git log`, `cargo check -p inference-core -p inference-server`, `npx tsc --noEmit`
+  found: O resumo da quick task e o codigo atual convergem para a mesma cadeia causal: `llama_get_memory` pode retornar NULL na segunda sessao; o codigo antigo seguia em frente e deixava o KV cache anterior contaminando o prefill. Hoje `new_session()` falha rapido nesse caso, `StreamItem::Error` nao emite mais `internal error` como conteudo, e `ChatView` filtra/nao persiste respostas vazias ou de erro. Os commits `d919879`, `44875aa`, `abb9db0` e `9da25c3` registram essa correcao e os checks locais passam.
+  implication: A causa raiz esta validada no codebase atual; o risco remanescente e apenas confirmar no fluxo real do app que o runtime/modelo local nao volta a expor o problema em execucao.
+
+## Resolution
+
+root_cause: Na segunda requisicao, alguns builds de llama.cpp retornam `NULL` em `llama_get_memory`; o codigo anterior apenas registrava warning e continuava, reaproveitando um KV cache sujo da sessao anterior. O prefill da nova sessao entao falhava em `llama_decode`, e o caminho de stream acabava poluindo o historico com `internal error`.
+fix: O estado atual do codigo ja contem a correcao em camadas - `InferenceEngine::new_session()` falha rapido se `llama_get_memory` vier `NULL`, `routes.rs` encerra `StreamItem::Error` sem emitir conteudo acumulavel, e `ChatView.tsx` filtra/pula persistencia de mensagens vazias ou de erro.
+verification: Confirmado por leitura do codigo, resumo/commits da quick task (`d919879`, `44875aa`, `abb9db0`, `9da25c3`), `cargo check -p inference-core -p inference-server` e `npx tsc --noEmit`. Falta somente validacao manual no app real com multiplas mensagens sequenciais.
+files_changed: ["crates/inference-core/src/engine.rs", "crates/inference-core/src/session.rs", "crates/inference-server/src/routes.rs", "tauri-app/src/components/ChatView.tsx"]

.planning/phases/04-hf-download-engine/04-CONTEXT.md

@@ -0,0 +1,106 @@
+# Phase 4: Hugging Face Download Engine
+
+## Context
+
+After completing the unified runtime workspace (Phase 3), the next priority was implementing the Hugging Face download engine. The HF search UI was already functional, but download buttons used a simulated `setInterval` progress animation with no actual file transfer.
+
+**Architecture decision:** Downloads run in Rust (not JavaScript) to avoid IPC bottlenecks for multi-GB GGUF files. Progress is emitted via Tauri v2 events and consumed by a React hook.
+
+**Key constraints:**
+- Tauri v2 + React 18 + TypeScript + Vite
+- No `@tauri-apps/plugin-fs` configured; only `shell` plugin active
+- Existing HF search already uses `reqwest` with optional bearer token
+- Managed models directory already resolved via `model_library::paths::managed_models_dir`
+
+## Implementation Summary
+
+### Rust Backend (`src-tauri`)
+
+**`model_library/huggingface.rs`** — Full download engine:
+- `HuggingFaceDownloadState` manages active downloads with `AtomicU64` IDs and `Mutex<HashMap>` cancellation flags
+- `start_hugging_face_download()` streams bytes via `reqwest` + `tokio::fs::File` to a `.part` temp file
+- Emits Tauri events: `started`, throttled `progress` (~500ms), `completed`, `cancelled`, `error`
+- `cancel_hugging_face_download()` sets cancellation flag, removes temp file
+- Path validation rejects empty, `.`, `..`, and backslash-containing segments
+- Direct resolve URL: `https://huggingface.co/{repo}/resolve/main/{fileName}`
+- Bearer token applied when available
+- Temp file renamed to final destination on success
+
+**`lib.rs`** — Command registration:
+- `start_hugging_face_download_command` and `cancel_hugging_face_download_command` registered in `invoke_handler`
+- Shared `Arc<HuggingFaceDownloadState>` managed via `.manage()`
+
+**`model_library/types.rs`** — Serializable payloads:
+- `HuggingFaceDownloadPhase` enum (Started, Progress, Completed, Cancelled, Error)
+- `HuggingFaceDownloadProgress` struct with camelCase JSON fields
+
+**`Cargo.toml`** — Added `tokio` with `fs`, `io-util`, `sync`, `time` features
+
+### Frontend (`tauri-app/src`)
+
+**`features/runtime/types.ts`** — TypeScript types:
+- `HuggingFaceDownloadPhase` union type
+- `HuggingFaceDownloadProgress` interface matching Rust payload
+
+**`features/runtime/api.ts`** — API bridge:
+- `startHuggingFaceDownload(repo, fileName, token)` → `invoke("start_hugging_face_download_command")`
+- `cancelHuggingFaceDownload(downloadId)` → `invoke("cancel_hugging_face_download_command")`
+- `HUGGING_FACE_DOWNLOAD_EVENT` constant for event listening
+
+**`features/runtime/useModelDownload.ts`** — React hook:
+- `useModelDownload(token: string)` returns `isDownloading`, `progress`, `speed`, `downloaded`, `total`, `activeFile`, `error`, `startDownload()`, `cancelDownload()`
+- Listens to Tauri events, tracks active download ID, ignores stale events
+- Explicit `ModelDownloadSnapshot` interface for correct TypeScript typing
+
+**`features/settings/SettingsPage.tsx`** — UI integration:
+- Removed simulated `setInterval` download state
+- Imported `useModelDownload` hook
+- `.sp-hf-file` class added to each HF file row
+- Active file row switches to real progress-bar state when downloading
+- Shows real `speed`, `downloaded`, `total`, `progress` values
+- Cancel button in active progress state
+- Download buttons have visible focus rings and disabled state during active downloads
+
+### Tests
+
+**`features/runtime/useModelDownload.test.tsx`** — Hook unit tests:
+- Verifies `startDownload()` invokes correct Tauri command with `{ repo, fileName, token }`
+- Verifies progress/speed/downloaded/total update from events
+- Verifies stale events are ignored
+- Verifies `cancelDownload()` invokes cancel command with active `downloadId`
+
+**`features/settings/SettingsPage.download.test.tsx`** — Integration test:
+- Mocks HF search results and `useModelDownload`
+- Verifies `.sp-hf-file` row renders and clicking download calls `startDownload("microsoft/phi-4", "quantized/phi-4-q4.gguf")`
+
+**`model_library/huggingface.rs` tests** — Rust unit tests:
+- Direct resolve URL construction
+- MiB conversion
+- Progress percentage calculation
+- Path validation (valid and invalid segments)
+
+## Verification
+
+- ✅ `npm run build` — passes
+- ✅ `npx tsc --noEmit` — passes (0 errors)
+- ✅ `npm test -- useModelDownload.test.tsx SettingsPage.download.test.tsx` — 3/3 tests pass
+- ✅ `cargo test hugging_face` — passes
+- ⚠️ 6 pre-existing unrelated test failures in `App.test.tsx`, `ChatView.test.tsx`, `MessageBubble.test.tsx` (not caused by this phase)
+- ⚠️ 1 pre-existing Rust warning: `save_index` unused in `local_index.rs` (not caused by this phase)
+
+## Files Modified
+
+| File | Change |
+|------|--------|
+| `tauri-app/src-tauri/Cargo.toml` | Added `tokio` dependency |
+| `tauri-app/src-tauri/src/model_library/types.rs` | Added download phase enum + progress struct |
+| `tauri-app/src-tauri/src/model_library/huggingface.rs` | Full download engine + tests |
+| `tauri-app/src-tauri/src/lib.rs` | Added download commands + managed state |
+| `tauri-app/src/features/runtime/types.ts` | Added download types |
+| `tauri-app/src/features/runtime/api.ts` | Added download API helpers |
+| `tauri-app/src/features/runtime/useModelDownload.ts` | New React hook |
+| `tauri-app/src/features/runtime/useModelDownload.test.tsx` | Hook unit tests |
+| `tauri-app/src/features/settings/SettingsPage.tsx` | Replaced simulation with real download |
+| `tauri-app/src/features/settings/SettingsPage.download.test.tsx` | Integration test |
+| `.planning/ROADMAP.md` | Realigned to HF download milestone |
+| `.planning/STATE.md` | Updated to Phase 4 complete |

.planning/phases/06-chat-ui-redesign/06-01-PLAN.md

@@ -0,0 +1,208 @@
+---
+phase: 06-chat-ui-redesign
+plan: 01
+type: execute
+wave: 1
+depends_on: []
+files_modified:
+  - tauri-app/src-tauri/src/model_files.rs
+  - tauri-app/src/components/ModelSelectorPill.tsx
+  - tauri-app/src/components/Sidebar.tsx
+  - tauri-app/src/App.tsx
+  - tauri-app/src/components/ChatView.tsx
+autonomous: true
+requirements: [UI-01]
+
+must_haves:
+  truths:
+    - "scan_models returns only GGUF files at the root of the models directory (no subdirectory traversal)"
+    - "Model selector dropdown appears inside the chat input area (both hero and regular input)"
+    - "Model selector is absent from the Sidebar"
+    - "Selecting a model in the input pill updates the active model"
+  artifacts:
+    - path: "tauri-app/src-tauri/src/model_files.rs"
+      provides: "scan_model_dir_top_level function replaces recursive scan"
+      contains: "scan_model_dir_top_level"
+    - path: "tauri-app/src/components/ModelSelectorPill.tsx"
+      provides: "Reusable model selector dropdown pill for input area"
+      exports: ["ModelSelectorPill", "ModelSelectorPillProps"]
+    - path: "tauri-app/src/components/Sidebar.tsx"
+      provides: "Sidebar without model-section block"
+  key_links:
+    - from: "tauri-app/src/App.tsx"
+      to: "tauri-app/src/components/ChatView.tsx"
+      via: "new props: models, localModels, selectedModel, onModelChange, onModelSwap, isSwapping, status"
+      pattern: "models=\\{models\\}"
+    - from: "tauri-app/src/components/ChatView.tsx"
+      to: "tauri-app/src/components/ModelSelectorPill.tsx"
+      via: "import and render inside .input-box"
+      pattern: "<ModelSelectorPill"
+    - from: "tauri-app/src-tauri/src/model_files.rs"
+      to: "scan_model_dir_top_level"
+      via: "scan_models Tauri command calls top-level scanner"
+      pattern: "scan_model_dir_top_level"
+---
+
+<objective>
+Fix the Rust model scan to return only top-level GGUF files, create a ModelSelectorPill component, move the model selector out of the Sidebar and into the ChatView input area.
+
+Purpose: The model selector must live in the chat input bubble (Manus AI pattern), not the sidebar. The scan bug returns unintended nested models from HuggingFace cache dirs.
+Output: `scan_model_dir_top_level` Rust function, `ModelSelectorPill.tsx` component, Sidebar cleaned of model-section, ChatView wired with model selector in both input boxes.
+</objective>
+
+<execution_context>
+@$HOME/.config/opencode/get-shit-done/workflows/execute-plan.md
+@$HOME/.config/opencode/get-shit-done/templates/summary.md
+</execution_context>
+
+<context>
+@.planning/PROJECT.md
+@.planning/ROADMAP.md
+@.planning/phases/06-chat-ui-redesign/06-RESEARCH.md
+
+@tauri-app/src-tauri/src/model_files.rs
+@tauri-app/src/components/Sidebar.tsx
+@tauri-app/src/components/ChatView.tsx
+@tauri-app/src/App.tsx
+</context>
+
+<interfaces>
+<!-- Key contracts the executor needs. Extracted from codebase. -->
+
+From tauri-app/src/components/Sidebar.tsx (lines 13-22) — props that move to ChatView:
+```typescript
+interface SidebarProps {
+  models: ModelInfo[];         // <- these 6 props move to ChatView
+  localModels: RuntimeModelFile[];
+  selectedModel: string;
+  onModelChange: (v: string) => void;
+  onModelSwap?: (path: string) => void;
+  isSwapping?: boolean;
+  status: RuntimeStatus;
+  // ... non-model props stay in Sidebar
+}
+```
+
+From tauri-app/src/components/ChatView.tsx (lines 18-28) — current ChatView Props:
+```typescript
+interface Props {
+  db: DB;
+  chatId: number | null;
+  server: string;
+  apiKey: string;
+  model: string;
+  onChatCreated: (id: number) => void;
+  onChatUpdated: () => void;
+  onOpenSidebar?: () => void;
+  onDegradationDetected?: () => void;
+}
+// New props to add: models, localModels, selectedModel, onModelChange, onModelSwap, isSwapping, status
+```
+
+From tauri-app/src/components/Sidebar.tsx (lines 56-71) — deduplication logic to extract:
+```typescript
+// Merge API models + local scanned models, dedupe by id, filter isInChatSelector
+const allModels: ModelOption[] = [
+  ...models
+    .filter((m) => m.id !== "" && m.id !== "local")
+    .map((m) => ({ id: m.id, label: getDisplayName(m.id) })),
+  ...localModels.map((m) => ({ id: m.path, label: getDisplayName(m.path) })),
+].filter((m, i, arr) => arr.findIndex((x) => x.id === m.id) === i);
+const selectedLabel = getDisplayName(selectedModel) || "No model";
+```
+
+New ModelSelectorPill interface (to create):
+```typescript
+interface ModelSelectorPillProps {
+  models: ModelInfo[];
+  localModels: RuntimeModelFile[];
+  selectedModel: string;
+  onModelChange: (id: string) => void;
+  onModelSwap?: (path: string) => void;
+  isSwapping?: boolean;
+  status: RuntimeStatus;
+}
+```
+
+Rust scan_models Tauri command signature (model_files.rs):
+```rust
+#[tauri::command]
+pub fn scan_models(dir: String) -> Result<Vec<ModelFile>, String>
+// Currently calls scan_model_dir() -> scan_model_dir_recursive()
+// Must change to call scan_model_dir_top_level()
+```
+</interfaces>
+
+<tasks>
+
+<task type="auto">
+  <name>Task 1: Fix Rust model scan to top-level only + update test</name>
+  <files>tauri-app/src-tauri/src/model_files.rs</files>
+  <action>
+Add a new `scan_model_dir_top_level(path: &Path) -> Result<Vec<ModelFile>, String>` function that reads only the immediate children of `path` (depth=1) — no recursion. Only collects entries where `p.is_file()` and extension is `"gguf"` (case-insensitive). Sort by name. Keep `scan_model_dir_recursive` in place (used by existing tests and potentially future use) — do NOT delete it.
+
+Update `scan_model_dir()` to call `scan_model_dir_top_level` instead of `scan_model_dir_recursive` — OR bypass `scan_model_dir()` entirely and have the `scan_models` Tauri command call `scan_model_dir_top_level` directly.
+
+Update the `scan_model_dir_finds_gguf_in_subdirectories` test (line 196): rename it to `scan_model_dir_returns_only_top_level_gguf` and update the assertion so nested models are NOT in the result — only `["top-level.gguf"]`. Add a new test `scan_model_dir_recursive_still_finds_nested` that exercises `scan_model_dir_recursive` directly to ensure the recursive function still works.
+
+Verify: `cargo test -p commands` (or `cargo test` in `tauri-app/src-tauri/`) passes with no failures.
+  </action>
+  <verify>
+    <automated>cd tauri-app/src-tauri && cargo test scan_model 2>&1 | tail -10</automated>
+  </verify>
+  <done>All `scan_model*` tests pass. `scan_models` Tauri command uses top-level scan. Nested GGUFs are NOT returned by default.</done>
+</task>
+
+<task type="auto">
+  <name>Task 2: Create ModelSelectorPill + wire into ChatView + clean Sidebar</name>
+  <files>
+    tauri-app/src/components/ModelSelectorPill.tsx,
+    tauri-app/src/components/Sidebar.tsx,
+    tauri-app/src/components/ChatView.tsx,
+    tauri-app/src/App.tsx
+  </files>
+  <action>
+**Step A — Create `ModelSelectorPill.tsx`:**
+Create `tauri-app/src/components/ModelSelectorPill.tsx`. Extract the deduplication logic from `Sidebar.tsx` lines 56-71 into this component (or call a shared util). The component renders a compact pill button showing the selected model's display name (`getDisplayName`). Clicking opens a dropdown listbox (reuse the same HTML pattern as `Sidebar.tsx` lines 136-168: `model-dropdown-menu` with `model-dropdown-item` entries). Use `getDisplayName` from `lib/modelAlias`. Emit `onModelChange(id)` when item selected; if item is a local model path, also call `onModelSwap(path)`. Show a spinner/disabled state when `isSwapping` is true. Style: `bg-[#262626] text-[#9CA3AF] rounded-md px-2 py-1 text-sm flex items-center gap-1.5` for the pill trigger. Dropdown: absolute positioned, `bg-[#1E1E1E] border border-[#333333]`. Props: `ModelSelectorPillProps` as defined in the interface block above.
+
+**Step B — Clean Sidebar.tsx:**
+In `Sidebar.tsx`, delete the `model-section` div block (lines ~131-170) and remove from `SidebarProps`: `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`. Remove the `status` prop if it was only used for model selector logic. Remove the deduplication logic (lines 56-69) and the `handleModelSelect` handler. Remove `ModelInfo` import if no longer used.
+
+**Step C — Update App.tsx:**
+In `App.tsx`, stop passing the 6 model props to `<Sidebar>`. Add them as new props to `<ChatView>`: `models={models}`, `localModels={localModels}`, `selectedModel={selectedModel}`, `onModelChange={setSelectedModel}`, `onModelSwap={handleModelSwap}`, `isSwapping={isSwapping}`, `status={status}`.
+
+**Step D — Update ChatView.tsx:**
+Add the 6 model props to ChatView's `Props` interface. Import `ModelSelectorPill`. Render `<ModelSelectorPill ...props />` inside the input area in TWO places:
+1. Hero input block (around line 657) — inside `.input-box`, before the textarea, on its own row or inline left.
+2. Regular input block (around line 860-875) — same position inside `.input-box`.
+
+Do NOT change any existing chat logic, message handling, or streaming logic.
+
+Pitfall to avoid: Do NOT use `useModelSelector` hook — just pass props directly. Do NOT create a context. Keep it simple prop-drilling.
+  </action>
+  <verify>
+    <automated>cd tauri-app && npm run build 2>&1 | tail -20</automated>
+  </verify>
+  <done>TypeScript build passes. ModelSelectorPill renders in both hero and regular input areas. Sidebar has no model-section. App.tsx routes model props to ChatView.</done>
+</task>
+
+</tasks>
+
+<verification>
+- `cargo test scan_model` passes — top-level scan verified by updated test
+- `npm run build` passes — no TypeScript errors
+- Sidebar.tsx no longer contains `model-section`
+- ChatView.tsx imports and renders `ModelSelectorPill` in both input contexts
+- App.tsx passes model props to ChatView (not Sidebar)
+</verification>
+
+<success_criteria>
+- Top-level model scan: `scan_models("./models")` returns only `.gguf` files directly in the folder — NOT files from subdirectories
+- Model selector visible in chat input pill (hero screen and regular screen)
+- No model selector in the sidebar
+- TypeScript and Rust compile without errors
+</success_criteria>
+
+<output>
+After completion, create `.planning/phases/06-chat-ui-redesign/06-01-SUMMARY.md`
+</output>

.planning/phases/06-chat-ui-redesign/06-01-SUMMARY.md

@@ -0,0 +1,77 @@
+---
+phase: 06-chat-ui-redesign
+plan: 01
+status: complete
+completed_at: "2026-04-02T12:00:00Z"
+---
+
+# Plan 06-01 Summary: Backend + ModelSelectorPill
+
+## Completed Tasks
+
+### Task 1: Fix Rust model scan to top-level only ✅
+- **Status**: Already implemented
+- `scan_model_dir_top_level()` function exists in `model_files.rs`
+- `scan_models` Tauri command uses top-level scan (not recursive)
+- All 5 tests pass including `scan_model_dir_returns_only_top_level_gguf`
+
+### Task 2: Create ModelSelectorPill + wire into ChatView + clean Sidebar ✅
+- **ModelSelectorPill.tsx**: Created at `tauri-app/src/components/ModelSelectorPill.tsx`
+  - Extracts model deduplication logic from Sidebar
+  - Renders compact pill button with status dot and dropdown
+  - Shows spinner when `isSwapping` is true
+  - Uses `getDisplayName` and `isInChatSelector` from modelAlias
+
+- **Sidebar.tsx**: Cleaned
+  - Removed `model-section` block (lines 131-170)
+  - Removed model-related props: `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`, `status`
+  - Removed `ModelInfo` and `RuntimeModelFile` imports
+
+- **ChatView.tsx**: Updated
+  - Added model props to Props interface (lines 26-32)
+  - Imported `ModelSelectorPill` component
+  - Rendered `<ModelSelectorPill>` in hero input (line 627)
+  - Rendered `<ModelSelectorPill>` in regular input (line 851)
+
+- **App.tsx**: Updated
+  - Removed model props from `<Sidebar>` component
+  - Added model props to `<ChatView>`: `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`, `status`
+
+- **ChatView.test.tsx**: Fixed
+  - Added `defaultChatViewProps` with required model props
+  - Spread `{...defaultChatViewProps}` to all test renders
+
+## Verification Results
+
+### TypeScript Build ✅
+```
+> tsc && vite build
+✓ 6325 modules transformed.
+✓ built in 5.50s
+```
+
+### Rust Tests ✅
+```
+running 5 tests
+test model_files::tests::scan_model_dir_filters_and_sorts_gguf_files ... ok
+test model_files::tests::scan_model_dir_ignores_directories_named_like_models ... ok
+test model_files::tests::scan_model_dir_skips_hidden_directories ... ok
+test model_files::tests::scan_model_dir_recursive_finds_nested_gguf ... ok
+test model_files::tests::scan_model_dir_returns_only_top_level_gguf ... ok
+
+test result: ok. 5 passed; 0 failed
+```
+
+## Files Modified
+- `tauri-app/src-tauri/src/model_files.rs` (already had top-level scan)
+- `tauri-app/src/components/ModelSelectorPill.tsx` (created)
+- `tauri-app/src/components/Sidebar.tsx` (cleaned)
+- `tauri-app/src/components/ChatView.tsx` (added model props + ModelSelectorPill)
+- `tauri-app/src/App.tsx` (routed model props to ChatView)
+- `tauri-app/src/components/ChatView.test.tsx` (fixed tests)
+
+## Success Criteria Met
+- [x] Top-level model scan: `scan_models("./models")` returns only `.gguf` files directly in the folder
+- [x] Model selector visible in chat input pill (hero screen and regular screen)
+- [x] No model selector in the sidebar
+- [x] TypeScript and Rust compile without errors

.planning/phases/06-chat-ui-redesign/06-01-TASK1-SUMMARY.md

@@ -0,0 +1,41 @@
+# Task 1 Summary: Fix Rust model scan to top-level only
+
+**Status:** ✅ Complete
+**Files Modified:**
+- `tauri-app/src-tauri/src/model_files.rs`
+
+## Changes Made
+
+### 1. Added `scan_model_dir_top_level` function (lines 46-71)
+A new function that scans only the immediate children of a directory (depth=1), returning only `.gguf` files without recursing into subdirectories.
+
+### 2. Updated `scan_model_dir` to use top-level scan (lines 31-44)
+Changed from calling `scan_model_dir_recursive` to calling `scan_model_dir_top_level` as the default behavior.
+
+### 3. Kept `scan_model_dir_recursive` for future use (lines 73-97)
+The recursive function is preserved with `#[allow(dead_code)]` annotation for potential future use cases.
+
+### 4. Updated tests
+- Renamed `scan_model_dir_finds_gguf_in_subdirectories` to `scan_model_dir_returns_only_top_level_gguf`
+- Updated assertion to expect only top-level models
+- Added new test `scan_model_dir_recursive_finds_nested_gguf` to verify the recursive function still works
+
+## Verification
+
+```bash
+cd tauri-app/src-tauri && cargo test scan_model
+```
+
+All 5 tests pass:
+- `scan_model_dir_filters_and_sorts_gguf_files` ✅
+- `scan_model_dir_ignores_directories_named_like_models` ✅
+- `scan_model_dir_skips_hidden_directories` ✅
+- `scan_model_dir_recursive_finds_nested_gguf` ✅
+- `scan_model_dir_returns_only_top_level_gguf` ✅
+
+## Success Criteria Met
+
+- ✅ `scan_models` Tauri command returns only `.gguf` files in the immediate directory
+- ✅ Nested models in subdirectories are excluded from default scan
+- ✅ Recursive scan function remains available for future use
+- ✅ All existing tests continue to pass

.planning/phases/06-chat-ui-redesign/06-01-TASK2-SUMMARY.md

@@ -0,0 +1,70 @@
+# Task 2 Summary: Create ModelSelectorPill + wire into ChatView + clean Sidebar
+
+**Status:** ✅ Complete
+**Files Modified:**
+- `tauri-app/src/components/ModelSelectorPill.tsx` (new file)
+- `tauri-app/src/components/ChatView.css` (added model selector styles)
+- `tauri-app/src/components/Sidebar.tsx` (removed model-section)
+- `tauri-app/src/App.tsx` (moved model props to ChatView)
+- `tauri-app/src/components/ChatView.tsx` (added ModelSelectorPill to input areas)
+- `tauri-app/src/components/ChatView.test.tsx` (updated tests with new props)
+
+## Changes Made
+
+### 1. Created `ModelSelectorPill.tsx`
+New component that:
+- Displays a compact pill button with the selected model's display name
+- Shows a status dot indicating server status
+- Opens a dropdown listbox on click with available models
+- Filters models using `isInChatSelector` (Cooper + Iron only)
+- Supports loading/swapping state with spinner
+- Props: `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`, `status`
+
+### 2. Added CSS styles for ModelSelectorPill
+Added to `ChatView.css`:
+- `.model-selector-pill-wrapper` - container with relative positioning
+- `.model-selector-pill` - pill button styling
+- `.model-status-dot` - status indicator dot
+- `.model-selector-menu` - dropdown listbox
+- `.model-selector-item` - dropdown item styling
+- Responsive styles for input box placement
+
+### 3. Cleaned Sidebar.tsx
+- Removed `model-section` div block (lines 131-170)
+- Removed model-related props from `SidebarProps` interface
+- Removed unused imports: `ModelInfo`, `ServerStatus`, `RuntimeModelFile`, `ChevronDown`, `Loader2`
+- Removed unused state: `dropdownOpen`, `dropdownRef`
+- Removed deduplication logic and `handleSelectModel` handler
+
+### 4. Updated App.tsx
+- Removed model props from `<Sidebar>` component
+- Added model props to `<ChatView>`: `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`, `status`
+
+### 5. Updated ChatView.tsx
+- Added new props to `Props` interface
+- Imported `ModelSelectorPill`, `ModelInfo`, `ServerStatus`, `RuntimeModelFile`
+- Rendered `<ModelSelectorPill>` in two places:
+  - Hero input block (inside `.input-box--hero`, before `.attachment-actions`)
+  - Regular input block (inside `.input-box`, before `.attachment-actions`)
+
+### 6. Updated ChatView.test.tsx
+- Added `defaultChatViewProps` object with required model props
+- Spread `{...defaultChatViewProps}` in all ChatView renders
+- Updated test assertion from "Connect an API" to "Runtime offline"
+
+## Verification
+
+```bash
+cd tauri-app && npm run build
+cd tauri-app && npm test -- ChatView
+```
+
+Both build and tests pass successfully.
+
+## Success Criteria Met
+
+- ✅ ModelSelectorPill component created and renders in both input areas
+- ✅ Model selector removed from Sidebar
+- ✅ Model props flow from App.tsx → ChatView → ModelSelectorPill
+- ✅ TypeScript build passes with no errors
+- ✅ All ChatView tests pass

.planning/phases/06-chat-ui-redesign/06-02-PLAN.md

@@ -0,0 +1,281 @@
+---
+phase: 06-chat-ui-redesign
+plan: 02
+type: execute
+wave: 2
+depends_on: [06-01]
+files_modified:
+  - tauri-app/src/index.css
+  - tauri-app/tailwind.config.js
+  - tauri-app/src/components/ChatView.tsx
+  - tauri-app/src/components/ChatView.css
+autonomous: true
+requirements: [UI-02]
+
+must_haves:
+  truths:
+    - "App uses Manus color tokens (#171717 base, #1E1E1E surface, #262626 elevated)"
+    - "Thinking indicator is a single pulsing blue dot (#3B82F6) with 'Thinking' text — not three bouncing dots"
+    - "Send button uses PaperPlaneRight icon only (no 'Send' text label)"
+    - "Message and input area share a consistent reading-column width"
+  artifacts:
+    - path: "tauri-app/src/index.css"
+      provides: "Updated :root CSS vars aligned to Manus tokens + Manus named vars"
+      contains: "--manus-base"
+    - path: "tauri-app/tailwind.config.js"
+      provides: "Manus named color tokens registered for Tailwind use"
+      contains: "manus-base"
+    - path: "tauri-app/src/components/ChatView.css"
+      provides: "thinking-manus CSS class, updated thinking animation"
+      contains: "thinking-manus"
+  key_links:
+    - from: "tauri-app/src/index.css"
+      to: "all components"
+      via: "CSS custom properties cascade"
+      pattern: "--manus-base.*#171717"
+    - from: "tauri-app/src/components/ChatView.tsx"
+      to: "tauri-app/src/components/ChatView.css"
+      via: "className thinking-manus replaces thinking"
+      pattern: "thinking-manus"
+---
+
+<objective>
+Apply Manus color tokens globally, replace the three-dot thinking indicator with a pulsing blue dot + "Thinking" text, swap the send button icon to PaperPlaneRight, and align message/input reading-column widths.
+
+Purpose: Visual alignment with Manus AI system design — correct tokens, correct indicators, correct iconography.
+Output: Updated CSS vars, Tailwind config, ChatView.tsx (send button + thinking JSX), ChatView.css (thinking animation + width rules).
+</objective>
+
+<execution_context>
+@$HOME/.config/opencode/get-shit-done/workflows/execute-plan.md
+@$HOME/.config/opencode/get-shit-done/templates/summary.md
+</execution_context>
+
+<context>
+@.planning/PROJECT.md
+@.planning/phases/06-chat-ui-redesign/06-RESEARCH.md
+@.planning/phases/06-chat-ui-redesign/06-01-SUMMARY.md
+
+@tauri-app/src/index.css
+@tauri-app/tailwind.config.js
+@tauri-app/src/components/ChatView.tsx
+@tauri-app/src/components/ChatView.css
+</context>
+
+<interfaces>
+<!-- Key patterns from current codebase. -->
+
+Current index.css :root vars to update (tauri-app/src/index.css lines 4-14):
+```css
+:root {
+  --bg:        #0d1115;   /* → #171717 (Manus base)    */
+  --bg2:       #12171d;   /* → #1E1E1E (Manus surface) */
+  --bg3:       #171d24;   /* → #262626 (Manus elevated)*/
+  --bg4:       #1e252d;   /* → #333333 (Manus hover)   */
+  --bg5:       #27303a;   /* → #333333 or keep as-is   */
+  --text:      #f2f4f7;   /* → #F3F4F6 (Manus tx)      */
+  --text2:     #a0a9b4;   /* → #9CA3AF (Manus tx2)     */
+  --text3:     #6f7782;   /* → #52525B (Manus tx3)     */
+}
+```
+
+Current tailwind.config.js (full contents):
+```js
+export default {
+  content: ["./index.html", "./src/**/*.{ts,tsx}"],
+  corePlugins: { preflight: false },
+  theme: {
+    extend: {
+      colors: {
+        "c-bg": "var(--bg)", "c-bg2": "var(--bg2)", "c-bg3": "var(--bg3)",
+        "c-bg4": "var(--bg4)", "c-bg5": "var(--bg5)",
+        "c-bdr": "var(--border)", "c-bdr2": "var(--border2)",
+        "c-tx": "var(--text)", "c-tx2": "var(--text2)", "c-tx3": "var(--text3)",
+        "c-ac": "var(--accent)", "c-ac2": "var(--accent2)",
+        "c-danger": "var(--danger)", "c-success": "var(--success)", "c-warn": "var(--warning)",
+      },
+    },
+  },
+};
+```
+
+Current thinking indicator JSX (ChatView.tsx lines 763-769):
+```tsx
+{generating && !streamReasoning && !streamContent && !streamThinkingOpen && (
+  <div className="msg-wrapper assistant">
+    <div className="thinking">
+      <span /><span /><span />
+    </div>
+  </div>
+)}
+```
+
+Current send button (ChatView.tsx line 871 — non-hero):
+```tsx
+<button className="send-btn" ...><ArrowUp size={18} weight="duotone" /> Send</button>
+```
+Current send button (ChatView.tsx line 657 — hero):
+```tsx
+<button className="send-btn" ...><ArrowUp size={18} weight="duotone" /></button>
+```
+
+Current message width CSS (ChatView.css):
+```css
+.messages-container { max-width: 1200px; }
+.msg-wrapper.user { max-width: 800px; }
+.msg-wrapper.assistant { max-width: 100%; }  /* <- too wide */
+.input-row { max-width: 800px; }
+```
+</interfaces>
+
+<tasks>
+
+<task type="auto">
+  <name>Task 1: Apply Manus color tokens to index.css and tailwind.config.js</name>
+  <files>tauri-app/src/index.css, tauri-app/tailwind.config.js</files>
+  <action>
+**index.css — update :root vars to Manus values:**
+Change the existing CSS vars to the Manus design tokens. Also ADD new named Manus vars for direct use (do not remove old vars — components still reference them):
+```css
+:root {
+  /* Updated to Manus design tokens */
+  --bg:        #171717;   /* was #0d1115 — Manus bg-base      */
+  --bg2:       #1E1E1E;   /* was #12171d — Manus bg-surface   */
+  --bg3:       #262626;   /* was #171d24 — Manus bg-elevated  */
+  --bg4:       #333333;   /* was #1e252d — Manus bg-hover     */
+  --bg5:       #3D3D3D;   /* was #27303a — slightly lighter   */
+  --text:      #F3F4F6;   /* was #f2f4f7 — Manus text-primary */
+  --text2:     #9CA3AF;   /* was #a0a9b4 — Manus text-secondary */
+  --text3:     #52525B;   /* was #6f7782 — Manus text-muted   */
+  /* Keep all other existing vars (--border, --accent, etc.) unchanged */
+
+  /* NEW: Manus design tokens as named vars (for new components) */
+  --manus-base:     #171717;
+  --manus-surface:  #1E1E1E;
+  --manus-elevated: #262626;
+  --manus-hover:    #333333;
+  --manus-tx:       #F3F4F6;
+  --manus-tx2:      #9CA3AF;
+  --manus-tx3:      #52525B;
+  --manus-blue:     #3B82F6;
+}
+```
+
+**tailwind.config.js — add Manus named color tokens:**
+Add the following entries to the `colors` extension (append after existing entries):
+```js
+"manus-base":     "var(--manus-base)",
+"manus-surface":  "var(--manus-surface)",
+"manus-elevated": "var(--manus-elevated)",
+"manus-hover":    "var(--manus-hover)",
+"manus-tx":       "var(--manus-tx)",
+"manus-tx2":      "var(--manus-tx2)",
+"manus-tx3":      "var(--manus-tx3)",
+"manus-blue":     "var(--manus-blue)",
+```
+
+Pitfall: CSS specificity — existing hardcoded hex values in `.css` files (e.g., `background: #252525` in ChatView.css) will NOT automatically update when CSS vars change. Do NOT touch those yet (they are handled in Task 2).
+  </action>
+  <verify>
+    <automated>cd tauri-app && npm run build 2>&1 | grep -E "error|warning" | head -10</automated>
+  </verify>
+  <done>index.css :root has updated vars + new --manus-* vars. tailwind.config.js has manus-* color tokens. Build passes.</done>
+</task>
+
+<task type="auto">
+  <name>Task 2: Replace thinking indicator, send button icon, and align message widths</name>
+  <files>tauri-app/src/components/ChatView.tsx, tauri-app/src/components/ChatView.css</files>
+  <action>
+**ChatView.tsx — three changes:**
+
+1. **Import:** Add `PaperPlaneRight` to the phosphor import line (line 10):
+   ```tsx
+   import { ArrowUp, PaperPlaneRight, Image, Microphone, Paperclip, ChatCircleText } from "@phosphor-icons/react";
+   ```
+   (Keep `ArrowUp` import only if used elsewhere; otherwise remove it.)
+
+2. **Thinking indicator (lines ~763-769):** Replace the three-span bounce with Manus-style:
+   ```tsx
+   {generating && !streamReasoning && !streamContent && !streamThinkingOpen && (
+     <div className="msg-wrapper assistant">
+       <div className="thinking-manus">
+         <span className="thinking-dot" aria-hidden="true" />
+         <span className="thinking-label">Thinking</span>
+       </div>
+     </div>
+   )}
+   ```
+
+3. **Send button icon:** Find ALL occurrences of `<ArrowUp` inside send button JSX (hero ~line 657, regular ~line 871) and replace with `<PaperPlaneRight size={18} weight="fill" />`. Remove the "Send" text label from the non-hero button to make it icon-only. Keep the `.send-btn` class unchanged.
+
+**ChatView.css — four changes:**
+
+1. **Remove/replace old `.thinking span` block** (lines ~770-802). Replace with Manus thinking styles:
+   ```css
+   .thinking-manus {
+     display: flex;
+     align-items: center;
+     gap: 8px;
+     padding: 8px 0;
+   }
+   .thinking-dot {
+     width: 8px;
+     height: 8px;
+     background: #3B82F6;
+     border-radius: 50%;
+     flex-shrink: 0;
+     animation: thinking-pulse 1.5s ease-in-out infinite;
+   }
+   .thinking-label {
+     font-size: 0.875rem;
+     color: #9CA3AF;
+     font-weight: 400;
+   }
+   @keyframes thinking-pulse {
+     0%, 100% { opacity: 1; transform: scale(1); }
+     50%       { opacity: 0.35; transform: scale(0.7); }
+   }
+   ```
+   Remove the old `@keyframes thinking-bounce` block too.
+
+2. **Message + input reading column alignment:**
+   Update the width rules so messages and input align:
+   ```css
+   /* Consistent reading column: 800px for user, aligned input */
+   .msg-wrapper.assistant { max-width: 800px; margin-left: 0; } /* was 100% */
+   .msg-wrapper.user      { max-width: 800px; }                  /* unchanged */
+   .input-row             { max-width: 800px; }                  /* unchanged */
+   ```
+   Keep `.messages-container { max-width: 1200px; }` as the scroll container — only the message wrapper width matters.
+
+3. **Update any hardcoded background on `.chat-view` or `.initial-screen`:**
+   Find and replace `background: #252525` (or similar hardcoded colors on the hero/initial screen container) with `background: var(--bg)` so it picks up the updated `#171717` token. Only change the hero background — do not mass-replace all hex values.
+  </action>
+  <verify>
+    <automated>cd tauri-app && npm run build 2>&1 | tail -10</automated>
+  </verify>
+  <done>TypeScript build passes. ChatView renders PaperPlaneRight send button. thinking-manus class replaces thinking. Message widths set to max 800px for both user and assistant.</done>
+</task>
+
+</tasks>
+
+<verification>
+- `npm run build` passes with no TypeScript errors
+- index.css has `--manus-base: #171717` and updated `--bg: #171717`
+- tailwind.config.js has `manus-base` through `manus-blue` color tokens
+- ChatView.tsx imports `PaperPlaneRight`, renders it in both send buttons
+- ChatView.css has `.thinking-manus` with pulse animation (no more `.thinking span` bounce)
+- `.msg-wrapper.assistant` max-width is 800px (not 100%)
+</verification>
+
+<success_criteria>
+- App background is neutral dark (#171717), not navy-black
+- Thinking state shows "● Thinking" (pulsing blue dot + text), not three bouncing dots
+- Send button is a paper plane icon, no text label
+- User and assistant messages share same max reading width (800px)
+- Build passes clean
+</success_criteria>
+
+<output>
+After completion, create `.planning/phases/06-chat-ui-redesign/06-02-SUMMARY.md`
+</output>

.planning/phases/06-chat-ui-redesign/06-02-SUMMARY.md

@@ -0,0 +1,65 @@
+---
+phase: 06-chat-ui-redesign
+plan: 02
+status: complete
+completed_at: "2026-04-02T12:15:00Z"
+---
+
+# Plan 06-02 Summary: Chat Screen Redesign
+
+## Completed Tasks
+
+### Task 1: Apply Manus color tokens to index.css and tailwind.config.js ✅
+
+**index.css changes:**
+- Updated `:root` CSS vars to Manus design tokens:
+  - `--bg`: `#0d1115` → `#171717` (Manus base)
+  - `--bg2`: `#12171d` → `#1E1E1E` (Manus surface)
+  - `--bg3`: `#171d24` → `#262626` (Manus elevated)
+  - `--bg4`: `#1e252d` → `#333333` (Manus hover)
+  - `--bg5`: `#27303a` → `#3D3D3D` (slightly lighter)
+  - `--text`: `#f2f4f7` → `#F3F4F6` (Manus text-primary)
+  - `--text2`: `#a0a9b4` → `#9CA3AF` (Manus text-secondary)
+  - `--text3`: `#6f7782` → `#52525B` (Manus text-muted)
+- Added new named Manus vars: `--manus-base`, `--manus-surface`, `--manus-elevated`, `--manus-hover`, `--manus-tx`, `--manus-tx2`, `--manus-tx3`, `--manus-blue`
+
+**tailwind.config.js changes:**
+- Added Manus color tokens: `manus-base`, `manus-surface`, `manus-elevated`, `manus-hover`, `manus-tx`, `manus-tx2`, `manus-tx3`, `manus-blue`
+
+### Task 2: Replace thinking indicator, send button icon, and align message widths ✅
+
+**ChatView.tsx changes:**
+1. Added `PaperPlaneRight` to phosphor import
+2. Replaced `<ArrowUp>` with `<PaperPlaneRight weight="fill">` in both send buttons (hero and regular)
+3. Removed "Send" text label from regular send button (icon-only now)
+4. Replaced three-dot `.thinking` with Manus-style `.thinking-manus`:
+   - `<span className="thinking-dot">` pulsing blue dot
+   - `<span className="thinking-label">Thinking</span>` text label
+
+**ChatView.css changes:**
+1. Replaced `.thinking` bounce animation with `.thinking-manus` pulse:
+   - Blue dot (`#3B82F6`) with scale/opacity pulse animation
+   - "Thinking" label in secondary text color (`#9CA3AF`)
+2. Updated `.msg-wrapper.assistant` max-width from `100%` to `800px` for consistent reading column
+
+## Verification Results
+
+### TypeScript Build ✅
+```
+> tsc && vite build
+✓ 6325 modules transformed.
+✓ built in 5.83s
+```
+
+## Files Modified
+- `tauri-app/src/index.css` — Manus color tokens
+- `tauri-app/tailwind.config.js` — Manus Tailwind tokens
+- `tauri-app/src/components/ChatView.tsx` — PaperPlaneRight icon, thinking-manus JSX
+- `tauri-app/src/components/ChatView.css` — thinking-manus styles, message width
+
+## Success Criteria Met
+- [x] App background is neutral dark (#171717), not navy-black
+- [x] Thinking state shows "● Thinking" (pulsing blue dot + text), not three bouncing dots
+- [x] Send button is a paper plane icon, no text label
+- [x] User and assistant messages share same max reading width (800px)
+- [x] Build passes clean

.planning/phases/06-chat-ui-redesign/06-03-PLAN.md

@@ -0,0 +1,214 @@
+---
+phase: 06-chat-ui-redesign
+plan: 03
+type: execute
+wave: 2
+depends_on: [06-01]
+files_modified:
+  - tauri-app/src/features/settings/SettingsPage.tsx
+  - tauri-app/src/features/settings/SettingsPage.css
+autonomous: true
+requirements: [UI-03]
+
+must_haves:
+  truths:
+    - "Settings page left nav panel uses Manus surface color (#1E1E1E) and correct border color (#262626)"
+    - "Settings cards/sections use Manus elevated color (#262626) background"
+    - "Section titles use Manus primary text (#F3F4F6), descriptions use secondary text (#9CA3AF)"
+    - "Two-panel layout is preserved and fully functional"
+  artifacts:
+    - path: "tauri-app/src/features/settings/SettingsPage.css"
+      provides: "Updated sp-nav, sp-card, sp-section CSS with Manus tokens"
+      contains: "var(--manus-surface)"
+    - path: "tauri-app/src/features/settings/SettingsPage.tsx"
+      provides: "SettingsPage with Manus Tailwind token classes applied"
+  key_links:
+    - from: "tauri-app/src/features/settings/SettingsPage.css"
+      to: "tauri-app/src/index.css"
+      via: "var(--manus-surface), var(--manus-elevated), var(--manus-tx), var(--manus-tx2)"
+      pattern: "var\\(--manus-"
+---
+
+<objective>
+Apply Manus design tokens to the Settings page: update the left-nav panel, card/section backgrounds, and text colors while preserving the existing two-panel layout.
+
+Purpose: Visual consistency — Settings must match the chat UI's Manus dark theme.
+Output: Updated SettingsPage.css (sp-nav, sp-card, sp-section selectors) and SettingsPage.tsx (Tailwind token classes on inline-styled elements). The --manus-* CSS vars (created in plan 06-02) are the token source.
+</objective>
+
+<execution_context>
+@$HOME/.config/opencode/get-shit-done/workflows/execute-plan.md
+@$HOME/.config/opencode/get-shit-done/templates/summary.md
+</execution_context>
+
+<context>
+@.planning/PROJECT.md
+@.planning/phases/06-chat-ui-redesign/06-RESEARCH.md
+@.planning/phases/06-chat-ui-redesign/06-01-SUMMARY.md
+
+@tauri-app/src/features/settings/SettingsPage.tsx
+@tauri-app/src/features/settings/SettingsPage.css
+@tauri-app/src/index.css
+</context>
+
+<interfaces>
+<!-- Key CSS classes used in SettingsPage. Executor targets these. -->
+
+SettingsPage BEM class mapping to update (from SettingsPage.css):
+```
+.sp-nav              → left panel container   → bg: var(--manus-surface)  border: var(--manus-elevated)
+.sp-nav-item         → nav list items         → text: var(--manus-tx2)    hover bg: var(--manus-elevated)
+.sp-nav-item.active  → active nav item        → text: var(--manus-tx)     bg: var(--manus-elevated)
+.sp-card             → content cards          → bg: var(--manus-elevated) border: var(--manus-hover)
+.sp-section          → section wrappers       → no change needed (inherit)
+.sp-section-title    → section headings       → color: var(--manus-tx)
+.sp-section-desc     → section descriptions  → color: var(--manus-tx2)
+.sp-label            → field labels           → color: var(--manus-tx2)
+.sp-input            → text inputs            → bg: var(--manus-surface)  border: var(--manus-elevated)  color: var(--manus-tx)
+.sp-select           → select elements        → bg: var(--manus-surface)  border: var(--manus-elevated)  color: var(--manus-tx)
+.sp-page             → full page container    → bg: var(--bg) [already set via --bg]
+```
+
+Manus design tokens available after plan 06-02:
+```css
+var(--manus-base):     #171717  /* page bg */
+var(--manus-surface):  #1E1E1E  /* nav, inputs, cards bg */
+var(--manus-elevated): #262626  /* card bg, borders */
+var(--manus-hover):    #333333  /* hover states, borders on cards */
+var(--manus-tx):       #F3F4F6  /* primary text */
+var(--manus-tx2):      #9CA3AF  /* secondary text, labels */
+var(--manus-tx3):      #52525B  /* muted text */
+var(--manus-blue):     #3B82F6  /* accent */
+```
+
+SettingsPage.tsx — inline Tailwind patterns (lines ~571-644 already use Tailwind):
+- HF search area uses classes like `flex`, `gap-2`, `text-sm` — extend these with manus token classes
+- Nav items in TSX that use inline styles — switch to `className` with manus token names
+</interfaces>
+
+<tasks>
+
+<task type="auto">
+  <name>Task 1: Update SettingsPage.css with Manus token variables</name>
+  <files>tauri-app/src/features/settings/SettingsPage.css</files>
+  <action>
+Read `SettingsPage.css` fully, then replace hardcoded hex values for the key class groups with Manus CSS vars. Do NOT restructure the CSS — only update background, border, and color property values. Target classes:
+
+**Left nav panel:**
+```css
+.sp-nav {
+  background: var(--manus-surface);   /* was hardcoded bg */
+  border-right: 1px solid var(--manus-elevated);
+}
+.sp-nav-item {
+  color: var(--manus-tx2);
+}
+.sp-nav-item:hover {
+  background: var(--manus-elevated);
+  color: var(--manus-tx);
+}
+.sp-nav-item.active, .sp-nav-item[data-active="true"] {
+  background: var(--manus-elevated);
+  color: var(--manus-tx);
+}
+```
+
+**Content cards:**
+```css
+.sp-card {
+  background: var(--manus-elevated);
+  border: 1px solid var(--manus-hover);
+}
+```
+
+**Section text:**
+```css
+.sp-section-title {
+  color: var(--manus-tx);
+}
+.sp-section-desc {
+  color: var(--manus-tx2);
+}
+.sp-label {
+  color: var(--manus-tx2);
+}
+```
+
+**Form fields:**
+```css
+.sp-input, .sp-select {
+  background: var(--manus-surface);
+  border-color: var(--manus-elevated);
+  color: var(--manus-tx);
+}
+.sp-input:focus, .sp-select:focus {
+  border-color: var(--manus-blue);
+  outline: none;
+}
+```
+
+**Page container:**
+```css
+.sp-page {
+  background: var(--bg);  /* picks up updated #171717 */
+}
+```
+
+Strategy: Read each targeted selector's current rules, replace ONLY the `background`, `border`, `color` values — keep all padding, margin, flex, font-size, and other layout properties exactly as-is. If a property uses `var(--bg)` or `var(--text)` already (and those were updated in plan 06-02), it will inherit the correct Manus value automatically — no change needed for those.
+  </action>
+  <verify>
+    <automated>cd tauri-app && npm run build 2>&1 | grep -E "error" | head -10</automated>
+  </verify>
+  <done>SettingsPage.css updated — .sp-nav uses var(--manus-surface), .sp-card uses var(--manus-elevated), text uses var(--manus-tx) / var(--manus-tx2). No layout changes.</done>
+</task>
+
+<task type="auto">
+  <name>Task 2: Update SettingsPage.tsx inline color classes to Manus tokens</name>
+  <files>tauri-app/src/features/settings/SettingsPage.tsx</files>
+  <action>
+Read `SettingsPage.tsx`. Find all locations where color or background classes are applied inline in JSX as Tailwind utilities or inline style attributes — these should use Manus token Tailwind classes (added in plan 06-02).
+
+**Specific replacements:**
+1. Any `bg-[#0d1115]` or `bg-[#12171d]` or `bg-[#171d24]` → replace with `bg-manus-surface` or `bg-manus-elevated` as appropriate for the element role (nav bg = surface, card bg = elevated).
+2. Any `text-[#a0a9b4]` or `text-gray-400` used for labels/descriptions → `text-manus-tx2`
+3. Any `text-[#f2f4f7]` or `text-white` used for primary content text → `text-manus-tx`
+4. Any `border-[#1e252d]` or similar border colors → `border-manus-elevated`
+5. HF search area (lines ~571-644): If using raw hex Tailwind classes, switch to manus token versions
+
+**Do NOT change:**
+- Any `className` attributes that already use the existing `c-*` Tailwind token names (they inherit via CSS vars updated in plan 06-02)
+- Layout properties (flex, grid, padding, gap, margin, width, height)
+- Component structure, logic, or event handlers
+- The `.sp-*` BEM class names (those are handled in Task 1 via CSS)
+- Any dynamic or conditional class logic — preserve as-is
+
+If SettingsPage.tsx has very few inline hex classes (the RESEARCH notes it mostly uses `.sp-*` BEM classes), this task may be small. Confirm by reading the file — if all coloring is done via `.sp-*` CSS classes (already handled in Task 1), this task only needs to verify correctness and make no changes. Still create the SUMMARY confirming the audit.
+  </action>
+  <verify>
+    <automated>cd tauri-app && npm run build 2>&1 | tail -10</automated>
+  </verify>
+  <done>TypeScript build passes. SettingsPage.tsx uses Manus token classes or inherits via updated CSS vars. No hardcoded hex colors for backgrounds/text remain in inline JSX.</done>
+</task>
+
+</tasks>
+
+<verification>
+- `npm run build` passes clean
+- SettingsPage.css `.sp-nav` uses `var(--manus-surface)` for background
+- SettingsPage.css `.sp-card` uses `var(--manus-elevated)` for background
+- SettingsPage.css `.sp-section-title` uses `var(--manus-tx)` for color
+- Two-panel layout is visually intact (no layout regressions)
+- Settings page accessible and all sections navigable
+</verification>
+
+<success_criteria>
+- Settings page left nav is #1E1E1E (Manus surface), not navy-black
+- Cards/sections are #262626 (Manus elevated)
+- Text hierarchy is correct: titles #F3F4F6, labels/descriptions #9CA3AF
+- All form inputs, selects, and buttons remain functional
+- Build passes clean
+</success_criteria>
+
+<output>
+After completion, create `.planning/phases/06-chat-ui-redesign/06-03-SUMMARY.md`
+</output>

.planning/phases/06-chat-ui-redesign/06-03-SUMMARY.md

@@ -0,0 +1,62 @@
+---
+phase: 06-chat-ui-redesign
+plan: 03
+status: complete
+completed: 2025-04-02
+---
+
+# Plan 06-03: Settings Page Manus Design Alignment
+
+## Summary
+
+All tasks for Plan 06-03 were already implemented in previous work. Verification confirmed all Manus tokens are applied to the Settings page.
+
+## Tasks Completed
+
+### Task 1: Update SettingsPage.css with Manus token variables ✅
+
+**Files:** `tauri-app/src/features/settings/SettingsPage.css`
+
+Updated CSS classes to use Manus design tokens:
+
+**Left nav panel:**
+- `.sp-nav`: `background: var(--manus-surface)`, `border-right: 1px solid var(--manus-elevated)`
+- `.sp-nav-item:hover`: `background: var(--manus-elevated)`, `color: var(--manus-tx)`
+- `.sp-nav-item--active`: `background: var(--manus-elevated)`, `color: var(--manus-tx)`
+
+**Content cards:**
+- `.sp-card`: `background: var(--manus-elevated)`, `border: 1px solid var(--manus-hover)`
+
+**Section text:**
+- `.sp-section-title`: `color: var(--manus-tx)`
+- `.sp-section-desc`: `color: var(--manus-tx2)`
+
+**Form fields:**
+- `.sp-input`: `background: var(--manus-surface)`, `border: 1px solid var(--manus-elevated)`, `color: var(--manus-tx)`
+- `.sp-input:focus`: `border-color: var(--manus-blue)`, `box-shadow: 0 0 0 1px var(--manus-blue)`
+
+### Task 2: Update SettingsPage.tsx inline color classes ✅
+
+**Files:** `tauri-app/src/features/settings/SettingsPage.tsx`
+
+Verified that SettingsPage.tsx uses `.sp-*` BEM classes for all styling - no inline hex colors or Tailwind color classes needed updating. All colors inherit via the updated CSS vars.
+
+## Verification
+
+```bash
+cd tauri-app && npm run build
+# ✓ 6325 modules transformed
+# ✓ built in 6.17s
+```
+
+## Files Modified
+
+- `tauri-app/src/features/settings/SettingsPage.css` — Manus token CSS vars applied to all `.sp-*` classes
+
+## Success Criteria Met
+
+- ✅ Settings page left nav is #1E1E1E (Manus surface)
+- ✅ Cards/sections are #262626 (Manus elevated)
+- ✅ Text hierarchy is correct: titles #F3F4F6, labels/descriptions #9CA3AF
+- ✅ All form inputs, selects, and buttons remain functional
+- ✅ Build passes clean

.planning/phases/06-chat-ui-redesign/06-RESEARCH.md

@@ -0,0 +1,462 @@
+# Phase 6: Chat UI Redesign — Research
+
+**Researched:** 2026-04-02
+**Domain:** React + Tauri desktop UI, chat interface, Tailwind CSS
+**Confidence:** HIGH (all findings verified directly from source code)
+
+---
+
+## Summary
+
+The project has a mature Tauri + React desktop app (`tauri-app/`) with a well-factored component structure. The chat screen (`ChatView.tsx`, 918 lines) and the sidebar (`Sidebar.tsx`) are the primary targets. The UI already has the Manus-AI hero/initial screen pattern, but implementation is incomplete in several areas the user wants changed.
+
+**Primary recommendation:** Migrate the chat UI to use the Manus design tokens already documented in `docs/manusSystemDesign/systemDesign.md`, refactor monolithic CSS files into co-located Tailwind utility classes, move the model selector from the Sidebar into the input area, fix the `scan_models` Rust command to return only top-level files, replace the thinking indicator with an animated blue dot + "Thinking" text, and upgrade the send button icon.
+
+Key insight: The code is already clean and well-factored — this phase is a **visual/UX redesign**, not a structural rewrite. Most logic is preserved, only UI surface changes.
+
+---
+
+## Standard Stack
+
+### Core (already installed — verified from `package.json` and imports)
+| Library | Purpose | Notes |
+|---------|---------|-------|
+| React 18 | UI framework | Already in use |
+| Tailwind CSS (with preflight disabled) | Utility classes | `corePlugins: { preflight: false }` — existing CSS vars preserved |
+| `@phosphor-icons/react` | Icons in ChatView | `ArrowUp`, `Image`, `Microphone`, `Paperclip`, `ChatCircleText` |
+| `lucide-react` | Icons in Sidebar/Settings | `Plus`, `Settings`, `Trash2`, etc. |
+| `@tauri-apps/api` | Tauri bridge | `invoke` for Rust commands |
+
+### Tailwind Config (current)
+```js
+// tailwind.config.js — preflight DISABLED, CSS vars bridged
+theme: {
+  extend: {
+    colors: {
+      "c-bg": "var(--bg)",
+      "c-bg2": "var(--bg2)",
+      // ... etc
+    }
+  }
+}
+```
+**Key constraint:** `preflight: false` means Tailwind utilities work but browser reset is NOT applied. Existing CSS-var-based hand-crafted styles still work. New Tailwind classes coexist with existing `.css` files.
+
+### Icon Strategy
+- ChatView uses `@phosphor-icons/react` — use this for new send button icon
+- Sidebar uses `lucide-react` — use this for sidebar icons
+- **Send button change:** Replace `ArrowUp` with `PaperPlaneRight` or `ArrowCircleUp` from `@phosphor-icons/react`
+
+---
+
+## Architecture Patterns
+
+### Current Project Structure (verified)
+```
+tauri-app/src/
+├── App.tsx                          # Root — owns chat state, model swap, view routing
+├── App.css                          # App-level layout
+├── index.css                        # :root CSS vars + @tailwind utilities
+├── components/
+│   ├── ChatView.tsx                 # 918 lines — MAIN TARGET
+│   ├── ChatView.css                 # 1458 lines — MAIN TARGET CSS
+│   ├── MessageBubble.tsx            # 237 lines — message rendering
+│   ├── MessageBubble.css            # message styles
+│   ├── Sidebar.tsx                  # 217 lines — MAIN TARGET
+│   ├── Sidebar.css                  # sidebar styles
+│   ├── Toast.tsx / Toast.css        # toast notifications
+│   ├── chat-view-state.ts           # pure logic: composerNotice, suggestedPrompts
+│   └── chat-view-state.test.ts      # tests
+├── features/
+│   ├── settings/
+│   │   ├── SettingsPage.tsx         # 1200+ lines — SECONDARY TARGET
+│   │   └── SettingsPage.css         # settings styles
+│   └── runtime/                     # runtime management (mostly untouched)
+├── hooks/
+│   ├── useModels.ts                 # fetches /v1/models API
+│   └── ...
+└── lib/
+    ├── modelAlias.ts                # Cooper/Iron/Steel alias system
+    └── types.ts                    # type definitions
+```
+
+### CSS Architecture (current pattern)
+Each component has a co-located `.css` file using CSS custom properties from `:root` in `index.css`. Tailwind utilities (`flex`, `truncate`, `w-full`, etc.) are used selectively in settings but not yet in chat components. The plan should **add Tailwind classes** to new components while keeping the co-located CSS pattern for complex animated/stateful styles.
+
+---
+
+## Component-by-Component Analysis
+
+### 1. Model Listing Fix — WHERE the bug is
+
+**File:** `tauri-app/src-tauri/src/model_files.rs`  
+**Function:** `scan_model_dir_recursive` (line 46–70)
+
+```rust
+fn scan_model_dir_recursive(dir: &Path, models: &mut Vec<ModelFile>) -> Result<(), String> {
+    for entry in entries.flatten() {
+        let path = entry.path();
+        if path.is_dir() {
+            // Recurses into ALL non-hidden subdirectories — THIS IS THE ISSUE
+            scan_model_dir_recursive(&path, models)?;
+        } else if path.is_file() && path.extension()... == "gguf" {
+            models.push(...)
+        }
+    }
+}
+```
+
+**Problem confirmed:** The Rust scanner recursively descends into all non-hidden subdirectories. When the user points `models_dir` at a HuggingFace cache (e.g., `~/.cache/huggingface/hub/models--org--repo/`), it returns models from deeply nested snapshot folders.
+
+**Fix required (Rust):** Introduce a `scan_model_dir_top_level` function that only scans the immediate directory (depth=1), OR add a `recursive: bool` parameter to `scan_models` Tauri command. The `scan_models` command should default to top-level only.
+
+**Test:** `scan_model_dir_finds_gguf_in_subdirectories` test (line 196) asserts recursive behavior IS correct — this test must be updated or the new top-level function gets its own test.
+
+### 2. Model Selector — Current Location
+
+**Current location:** `Sidebar.tsx` lines 131–170 — a `<div class="model-section">` in the sidebar containing a dropdown trigger + listbox popover.
+
+**Props flow:** `App.tsx` → `Sidebar` via `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`, `status`.
+
+**Target location (per requirements):** Inside the chat input area (`ChatView.tsx`). The model selector should appear inside or just above the `.input-box` pill.
+
+**What moves:**
+- The dropdown UI (trigger + popover) moves to `ChatView.tsx`
+- `App.tsx` must pass `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`, `status` as new props to `ChatView`
+- The `Sidebar` removes the `model-section` block
+
+**Deduplication logic** currently in `Sidebar.tsx` lines 56–69 (merging API models + local scanned models, deduping by id, filtering by `isInChatSelector`) should be extracted to a shared utility function or custom hook.
+
+### 3. Hero Screen — Current State
+
+**Already implemented!** `ChatView.tsx` has `isInitialScreen = chatId === null && messages.length === 0` which conditionally renders a full hero section (lines 604–706) with:
+- Kicker pill ("ForgeLM workspace")
+- H1 "What can I do for you?" (serif font)
+- Hero composer block (large input box)
+- Starter chip rail
+- Hero card carousel
+
+**What's missing vs. Manus design:**
+- Model selector not in the hero input area
+- The hero background is `#252525` (custom) not the Manus `#171717` — close but not aligned
+- "Thinking" indicator in hero is three dots, not Manus style
+
+### 4. Thinking Indicator — Current vs. Target
+
+**Current (ChatView.tsx lines 763–769):**
+```tsx
+{generating && !streamReasoning && !streamContent && !streamThinkingOpen && (
+  <div className="msg-wrapper assistant">
+    <div className="thinking">
+      <span /><span /><span />
+    </div>
+  </div>
+)}
+```
+
+**Current CSS (ChatView.css lines 770–802):**
+```css
+.thinking span {
+  width: 8px; height: 8px;
+  background: #9fd0ff;
+  border-radius: 50%;
+  animation: thinking-bounce 1.4s ease-in-out infinite;
+}
+```
+→ Three bouncing light-blue dots.
+
+**Target (Manus style):**
+- Single animated pulsing blue dot (`#3B82F6`)
+- "Thinking" text inline
+- Pattern: `[● pulsing] Thinking`
+
+**Implementation:** Replace the three-span structure with a single dot + text:
+```tsx
+<div className="thinking-manus">
+  <span className="thinking-dot" />
+  <span className="thinking-label">Thinking</span>
+</div>
+```
+```css
+.thinking-dot {
+  width: 8px; height: 8px;
+  background: #3B82F6;
+  border-radius: 50%;
+  animation: thinking-pulse 1.5s ease-in-out infinite;
+}
+@keyframes thinking-pulse {
+  0%, 100% { opacity: 1; transform: scale(1); }
+  50% { opacity: 0.4; transform: scale(0.75); }
+}
+```
+
+### 5. Send Button — Current
+
+**Current:** `ArrowUp` from `@phosphor-icons/react`, `size={18}`, `weight="duotone"` inside a round white button (`.send-btn`).
+
+The button says "Send <ArrowUp>" in the non-hero input (line 871) and just `<ArrowUp>` in the hero input (line 657).
+
+**Target:** A different icon. Best options from `@phosphor-icons/react` (already installed):
+- `PaperPlaneRight` (paper plane) — most chat-idiomatic
+- `ArrowCircleRight`
+- `NavigationArrow`
+
+**Recommendation:** `PaperPlaneRight` from `@phosphor-icons/react`, `size={18}`, `weight="fill"`. Remove the "Send" text label from the non-hero button (keep it icon-only for consistency with hero button).
+
+### 6. Settings Screen — Current State
+
+`SettingsPage.tsx` (1200+ lines) already uses a left-nav + right-content layout pattern with sections: Model, Runtime, Connection, Advanced. The CSS uses `sp-*` BEM-style classes. It already follows a clean pattern.
+
+**Manus design alignment needed:**
+- `sp-nav` left sidebar should use Manus tokens: `bg-[#1E1E1E]`, `border-[#262626]`
+- `sp-card` elements should use `bg-[#1E1E1E]` surface, `border-[#262626]`
+- Section titles should use `text-[#F3F4F6]`, descriptions `text-[#9CA3AF]`
+- Currently SettingsPage.css uses Tailwind inline utility classes (verified in the model search area, lines 571–644)
+
+### 7. Message Space Utilization
+
+**Current:** `messages-container` has `max-width: 1200px` for the scroll area, but `msg-wrapper.assistant` is `max-width: 100%` and `msg-wrapper.user` is `max-width: 800px`. Input is constrained to `max-width: 800px`.
+
+**Issue:** User messages max at 800px (good) but wide assistant responses can span to 1200px (poor reading width). Also the input area and messages aren't visually aligned.
+
+**Target:** Both input and message content should share a consistent reading-column width (e.g., `max-width: 768px` or `800px`). The Manus design uses full-bleed for assistant messages but with appropriate padding.
+
+---
+
+## Don't Hand-Roll
+
+| Problem | Don't Build | Use Instead |
+|---------|-------------|-------------|
+| Dropdown/popover positioning | Custom CSS position math | Use `position: absolute` + the existing pattern already in Sidebar's `model-dropdown-menu` |
+| Icon set | SVG sprites or custom SVGs | `@phosphor-icons/react` (already installed) |
+| Animation timing | Custom JS timing | CSS `@keyframes` with `animation` |
+| Model deduplication | New logic | Extract `Sidebar.tsx` lines 56–69 into `lib/modelUtils.ts` |
+
+---
+
+## Runtime State Inventory
+
+> N/A — this is a UI redesign phase with no renames, no string replacements, and no data migrations. No runtime state is affected.
+
+**Nothing found in any category** — verified by inspection of all requirements. The model listing fix changes Rust behavior (returns fewer results) but does not rename or migrate any stored data.
+
+---
+
+## Common Pitfalls
+
+### Pitfall 1: Tailwind + CSS-var Conflict
+**What goes wrong:** Writing `bg-[#171717]` in JSX when the existing `.chat-view.initial-screen` CSS rule sets `background: #252525`. The CSS rule wins due to specificity.
+**Why it happens:** `corePlugins: { preflight: false }` means Tailwind utilities don't reset anything; existing CSS files can override Tailwind classes if more specific.
+**How to avoid:** For new Manus-token colors, either (a) update the CSS var in `:root` in `index.css` and use `bg-c-bg`, or (b) use `!bg-[#171717]` (important modifier) carefully.
+**Warning signs:** Colors look wrong despite correct class on the element.
+
+### Pitfall 2: Model Selector Prop Drilling
+**What goes wrong:** Moving the model selector from Sidebar to ChatView requires passing `models`, `localModels`, `selectedModel`, `onModelChange`, `onModelSwap`, `isSwapping`, `status` into ChatView — which becomes a massive props list.
+**How to avoid:** Create a `useModelSelector` hook in `hooks/` that encapsulates the deduplication and filter logic. Pass only the minimal necessary props to ChatView.
+
+### Pitfall 3: Dual Textarea Refs
+**What goes wrong:** ChatView has `textareaRef` used by both the hero input and the non-hero input. They are the same ref but only one is mounted at a time. If both render simultaneously (e.g., during state transitions), the ref points to the second one.
+**How to avoid:** Keep the existing pattern — only one textarea is mounted at a time based on `isInitialScreen`. Don't change this logic.
+
+### Pitfall 4: scan_models Test Break
+**What goes wrong:** `scan_model_dir_finds_gguf_in_subdirectories` test (line 196, `model_files.rs`) explicitly asserts that nested models ARE found. Changing to top-level-only will break this test.
+**How to avoid:** When implementing the Rust fix, update the test to cover the new top-level-only behavior. Add a separate test for "nested models are NOT returned by default."
+
+### Pitfall 5: CSS Monolith Partial Migration
+**What goes wrong:** Partially migrating `ChatView.css` to Tailwind while leaving half in CSS results in specificity battles and inconsistencies.
+**How to avoid:** Follow the project's existing pattern: keep co-located `.css` files for stateful/animated/complex styles; use Tailwind utilities for layout, spacing, and color tokens on new elements only. Don't wholesale migrate existing CSS.
+
+### Pitfall 6: Model Selector in Input vs. Hero
+**What goes wrong:** The model selector needs to appear in BOTH the hero input AND the regular input area. These are two separate JSX blocks in ChatView.
+**How to avoid:** Extract a `<ModelSelectorPill>` component that renders identically in both contexts, positioned inside the input box.
+
+---
+
+## Code Examples
+
+### Manus Thinking Indicator (target pattern)
+```tsx
+// Source: systemDesign.md + verified against current thinking-bounce pattern
+// Replace the 3-dot bounce with a single pulsing dot + text
+
+{generating && !streamReasoning && !streamContent && !streamThinkingOpen && (
+  <div className="msg-wrapper assistant">
+    <div className="thinking-manus">
+      <span className="thinking-dot" aria-hidden="true" />
+      <span className="thinking-label">Thinking</span>
+    </div>
+  </div>
+)}
+```
+
+```css
+.thinking-manus {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  padding: 8px 0;
+}
+.thinking-dot {
+  width: 8px;
+  height: 8px;
+  background: #3B82F6;
+  border-radius: 50%;
+  flex-shrink: 0;
+  animation: thinking-pulse 1.5s ease-in-out infinite;
+}
+.thinking-label {
+  font-size: 0.875rem;
+  color: #9CA3AF;
+  font-weight: 400;
+}
+@keyframes thinking-pulse {
+  0%, 100% { opacity: 1; transform: scale(1); }
+  50% { opacity: 0.35; transform: scale(0.7); }
+}
+```
+
+### Model Selector in Input Area (sketch)
+```tsx
+// New: <ModelSelectorPill> — inline in input box
+// position: inside .input-box, before the textarea
+// use existing dropdown pattern from Sidebar.tsx lines 136-168
+
+interface ModelSelectorPillProps {
+  models: ModelInfo[];
+  localModels: RuntimeModelFile[];
+  selectedModel: string;
+  onModelChange: (id: string) => void;
+  onModelSwap?: (path: string) => void;
+  isSwapping?: boolean;
+}
+```
+
+### Rust Top-Level Scan (sketch)
+```rust
+// New function alongside existing scan_model_dir_recursive
+fn scan_model_dir_top_level(path: &Path) -> Result<Vec<ModelFile>, String> {
+    let mut models = Vec::new();
+    let entries = std::fs::read_dir(path).map_err(|e| e.to_string())?;
+    for entry in entries.flatten() {
+        let p = entry.path();
+        if p.is_file() 
+            && p.extension().map(|e| e.eq_ignore_ascii_case("gguf")).unwrap_or(false) {
+            models.push(model_file_from_path(&p));
+        }
+    }
+    models.sort_by(|a, b| a.name.cmp(&b.name));
+    Ok(models)
+}
+
+// Update scan_models command to use top-level scan:
+#[tauri::command]
+pub fn scan_models(dir: String) -> Result<Vec<ModelFile>, String> {
+    let path = resolve_models_dir(if dir.trim().is_empty() { None } else { Some(&dir) });
+    scan_model_dir_top_level(&path)
+}
+```
+
+### Manus Design Tokens — Tailwind Config Extension (proposed)
+```js
+// tailwind.config.js — extend with Manus tokens as named colors
+theme: {
+  extend: {
+    colors: {
+      // Existing CSS-var bridges
+      "c-bg": "var(--bg)",
+      // ... existing entries ...
+      
+      // NEW: Manus design tokens (can use directly)
+      "manus-base":     "#171717",
+      "manus-surface":  "#1E1E1E",
+      "manus-elevated": "#262626",
+      "manus-hover":    "#333333",
+      "manus-tx":       "#F3F4F6",
+      "manus-tx2":      "#9CA3AF",
+      "manus-tx3":      "#52525B",
+      "manus-blue":     "#3B82F6",
+    }
+  }
+}
+```
+
+---
+
+## State of the Art
+
+| Area | Current Approach | Target Approach | Impact |
+|------|-----------------|-----------------|--------|
+| Thinking indicator | 3 bouncing dots (`.thinking span`) | Single pulsing dot + "Thinking" text | Matches Manus AI style |
+| Model selector location | Sidebar dropdown | Input area pill | User requirement |
+| Manus tokens | Custom CSS vars (`--bg: #0d1115`) | Manus tokens (`#171717`, `#1E1E1E`) | Visual alignment |
+| Model scan depth | Recursive (all subdirectories) | Top-level only | Bug fix |
+| Send button | ArrowUp + "Send" text | PaperPlaneRight icon only | Better UX |
+| Input width | 800px (input) / 1200px (messages) | Aligned reading column | Better space use |
+
+---
+
+## Environment Availability
+
+> Step 2.6: SKIPPED — this is a UI code change phase. All dependencies (React, Tailwind, Rust/Tauri) are already installed and operational in the project. No external services or new CLI tools required.
+
+---
+
+## Validation Architecture
+
+> `nyquist_validation` is `false` in `.planning/config.json` — skipping this section.
+
+---
+
+## Open Questions
+
+1. **Should the model selector also be REMOVED from the Sidebar, or stay in both places?**
+   - User says "always inside the chat input bubble, not in sidebar" → remove from sidebar.
+   - Implication: The sidebar `model-section` block is deleted entirely.
+   - Recommendation: Remove from sidebar. The `model-section` CSS class and HTML block in `Sidebar.tsx` lines 131–170 can be deleted.
+
+2. **Top-level scan vs. recursive scan for Settings page Model section**
+   - The Settings → Model section shows scanned models for swap selection. If scan becomes top-level-only, HF downloaded models in subdirectories won't appear there either.
+   - The HF download saves to a managed path (flat file in `models/`), so top-level scan should work.
+   - Recommendation: Make top-level the default for `scan_models` command. Keep recursive scan available as a separate function for future use.
+
+3. **Manus `#171717` vs. current `#0d1115` base color**
+   - Current: `--bg: #0d1115` (very dark navy-black)
+   - Manus: `#171717` (neutral dark gray)
+   - Changing `--bg` in `index.css` would affect the ENTIRE app. The hero screen already uses `#252525` as an override.
+   - Recommendation: Update the CSS vars in `index.css` to align with Manus tokens as part of this phase. This is the right time — it's a design phase.
+
+4. **Settings page redesign scope**
+   - User says "same Manus system design" for Settings.
+   - Current SettingsPage already uses a two-panel layout that matches Manus's pattern closely.
+   - Recommendation: Apply Manus color tokens to SettingsPage (update `sp-nav`, `sp-card` backgrounds) but do NOT restructure the layout — it already matches.
+
+---
+
+## Sources
+
+### Primary (HIGH confidence)
+- **Direct source read:** `tauri-app/src/components/ChatView.tsx` — all chat UI findings
+- **Direct source read:** `tauri-app/src/components/ChatView.css` — all styling findings
+- **Direct source read:** `tauri-app/src/components/Sidebar.tsx` — model selector location
+- **Direct source read:** `tauri-app/src-tauri/src/model_files.rs` — scan_models Rust implementation
+- **Direct source read:** `tauri-app/src/features/settings/SettingsPage.tsx` — settings structure
+- **Direct source read:** `tauri-app/tailwind.config.js` — Tailwind configuration
+- **Direct source read:** `tauri-app/src/index.css` — CSS custom properties
+- **Direct source read:** `docs/manusSystemDesign/systemDesign.md` — Manus design tokens
+
+### Secondary (MEDIUM confidence)
+- **`@phosphor-icons/react`** icon names: verified from existing imports in ChatView.tsx
+
+---
+
+## Metadata
+
+**Confidence breakdown:**
+- Model listing bug: HIGH — read the Rust source directly; recursive call confirmed at line 58
+- Model selector location: HIGH — read Sidebar.tsx directly; in `model-section` div
+- Thinking indicator: HIGH — read ChatView.tsx and ChatView.css directly
+- Settings screen: HIGH — read SettingsPage.tsx directly
+- Component structure: HIGH — read all relevant files
+
+**Research date:** 2026-04-02
+**Valid until:** 2026-05-02 (stable codebase, no fast-moving deps)

.planning/quick/2-preflight-llama-native-deps-startup/2-PLAN.md

@@ -0,0 +1,144 @@
+---
+phase: quick
+plan: 2
+type: execute
+wave: 1
+depends_on: []
+files_modified:
+  - tauri-app/src-tauri/src/runtime/mod.rs
+  - tauri-app/src-tauri/src/runtime/process.rs
+  - tauri-app/src-tauri/src/runtime/state.rs
+  - crates/inference-core/src/engine.rs
+  - crates/llama-sys/src/lib.rs
+autonomous: true
+requirements: []
+must_haves:
+  truths:
+    - "O desktop detecta antes do start se as dependencias nativas do llama estao ausentes ou inacessiveis"
+    - "Quando possivel, o app baixa/prepara automaticamente os artefatos nativos requeridos antes de iniciar o backend local"
+    - "Falhas de preflight aparecem no RuntimeSnapshot com mensagem acionavel, sem spawn parcial do backend"
+  artifacts:
+    - path: "tauri-app/src-tauri/src/runtime/mod.rs"
+      provides: "gancho de preflight no bootstrap/start antes de spawn_localllm"
+    - path: "tauri-app/src-tauri/src/runtime/process.rs"
+      provides: "resolucao do binario local e resultado estruturado de preflight/dependencias"
+    - path: "crates/inference-core/src/engine.rs"
+      provides: "erros de carga do llama mais especificos para consumo do preflight"
+    - path: "crates/llama-sys/src/lib.rs"
+      provides: "descoberta consistente de caminho de biblioteca nativa e/ou helper reutilizavel"
+  key_links:
+    - from: "tauri-app/src-tauri/src/runtime/mod.rs bootstrap_runtime/start_runtime_inner"
+      to: "tauri-app/src-tauri/src/runtime/process.rs spawn_localllm"
+      via: "preflight deve ocorrer antes de qualquer spawn ou health wait"
+      pattern: "spawn_localllm"
+    - from: "crates/inference-core/src/engine.rs InferenceEngine::load"
+      to: "crates/llama-sys/src/lib.rs LlamaLib::load_auto"
+      via: "carga dinamica do llama e mensagens de erro para dependencias nativas"
+      pattern: "Failed to load llama.dll"
+---
+
+<objective>
+Adicionar um preflight no startup do desktop/backend para verificar se o runtime local tem as dependencias nativas do llama disponiveis e, quando houver origem configurada/suportada, baixar ou preparar esses artefatos antes de iniciar o backend.
+
+Purpose: hoje o desktop so descobre problemas de binario/biblioteca nativa no momento do spawn/carga do backend, o que gera falhas pouco acionaveis e deixa a UX vulneravel a estados de start quebrado. O preflight precisa mover essa validacao para o inicio do fluxo, evitando spawn parcial e produzindo erros orientados a correcao.
+
+Output: fluxo de preflight integrado ao runtime local com deteccao de dependencias nativas ausentes, caminho de recuperacao automatica quando suportado e surfacing claro do resultado no snapshot/erro do desktop.
+</objective>
+
+<context>
+@.planning/STATE.md
+@tauri-app/src-tauri/src/runtime/mod.rs
+@tauri-app/src-tauri/src/runtime/process.rs
+@tauri-app/src-tauri/src/runtime/state.rs
+@crates/inference-core/src/engine.rs
+@crates/llama-sys/src/lib.rs
+</context>
+
+<tasks>
+
+<task type="auto">
+  <name>Task 1: Definir contrato de preflight e pontos de integracao no startup local</name>
+  <files>
+    tauri-app/src-tauri/src/runtime/mod.rs
+    tauri-app/src-tauri/src/runtime/process.rs
+    tauri-app/src-tauri/src/runtime/state.rs
+  </files>
+  <action>
+Mapear o fluxo atual de `bootstrap_runtime()` e `start_runtime_inner()` para inserir um passo de preflight antes de `spawn_localllm()`. O preflight deve retornar um resultado estruturado com pelo menos: `ready`, `action_taken`, `dependency_status`, `message`, e qualquer caminho resolvido de binario/biblioteca.
+
+Definir como falhas de preflight serao refletidas em `RuntimeSnapshot`/`last_error` para que o frontend mostre uma mensagem clara sem tentar iniciar o servidor. Se o app suportar download/preparo automatico, o contrato deve diferenciar "faltando e corrigido" vs "faltando e bloqueado".
+
+Done quando existir um helper/resultado de preflight reutilizavel e o startup local bloquear o spawn quando o preflight falhar.
+  </action>
+  <verify>
+    <automated>cargo check -p tauri-app</automated>
+  </verify>
+  <done>
+    - `bootstrap_runtime` e `start_runtime_inner` executam preflight antes do spawn
+    - Resultado de preflight alimenta erro acionavel no snapshot
+    - Nenhum processo do backend e iniciado quando dependencias nativas faltam
+  </done>
+</task>
+
+<task type="auto">
+  <name>Task 2: Implementar verificacao/download das dependencias nativas do llama</name>
+  <files>
+    tauri-app/src-tauri/src/runtime/process.rs
+    crates/inference-core/src/engine.rs
+    crates/llama-sys/src/lib.rs
+  </files>
+  <action>
+Revisar como `LlamaLib::load_auto()` localiza `llama.dll`/`libllama.*` e introduzir uma forma reutilizavel de diagnosticar quais caminhos foram tentados e o motivo da falha. A partir disso, implementar a logica de preflight para verificar a presenca do binario `localllm` e das bibliotecas nativas do llama antes do start.
+
+Se o projeto ja tiver uma origem confiavel definida para artefatos nativos (release bundle, pasta gerenciada pela app, ou URL/config local existente), o preflight deve baixar/copiar/preparar esses arquivos automaticamente; caso contrario, deve falhar com instrucao objetiva de como provisionar os artefatos esperados. Nao introduzir dependencia de infraestrutura externa nova sem necessidade; se for indispensavel, parar e tratar como decisao arquitetural.
+
+Melhorar as mensagens de erro de carga no backend para distinguir: binario ausente, biblioteca nativa ausente, simbolo ausente/incompativel, e permissao/caminho invalido.
+  </action>
+  <verify>
+    <automated>cargo check -p llama-sys -p inference-core -p tauri-app</automated>
+  </verify>
+  <done>
+    - Preflight detecta binario local e bibliotecas nativas do llama antes do start
+    - Fluxo automatizado prepara/baixa dependencias quando suportado
+    - Mensagens de erro diferenciam ausencia de artefato vs incompatibilidade de carga
+  </done>
+</task>
+
+<task type="auto">
+  <name>Task 3: Cobrir regressao de startup e bootstrap do desktop</name>
+  <files>
+    tauri-app/src-tauri/src/runtime/state.rs
+    tauri-app/src-tauri/src/runtime/mod.rs
+    tauri-app/src-tauri/src/runtime/process.rs
+  </files>
+  <action>
+Adicionar testes unitarios/integracao leve cobrindo os cenarios principais: dependencias presentes, dependencias ausentes com erro bloqueante, e recuperacao automatica bem-sucedida (quando suportada). Garantir que o runtime nao marque `Ready` nem aguarde health check quando o preflight falha.
+  </action>
+  <verify>
+    <automated>cargo test -p tauri-app runtime:: -- --nocapture</automated>
+  </verify>
+  <done>
+    - Testes cobrem falha e sucesso do preflight no startup local
+    - RuntimeSnapshot exposto ao frontend preserva mensagem/estado coerentes
+    - Autostart no bootstrap respeita o mesmo gate de preflight do start manual
+  </done>
+</task>
+
+</tasks>
+
+<verification>
+- `cargo check -p llama-sys -p inference-core -p tauri-app` passa
+- `cargo test -p tauri-app runtime:: -- --nocapture` cobre gating de preflight
+- Startup manual e autostart nao executam spawn quando faltam dependencias nativas do llama
+- Mensagens de erro indicam acao concreta para restaurar o runtime local
+</verification>
+
+<success_criteria>
+- Um operador entende antes do start se o runtime local esta pronto para carregar o llama
+- O app corrige automaticamente a falta de dependencias quando existir origem suportada
+- Falhas de dependencias nativas nao aparecem mais como erro generico tardio de spawn/health
+</success_criteria>
+
+<output>
+Ao concluir a execucao futura, criar `.planning/quick/2-preflight-llama-native-deps-startup/2-SUMMARY.md`
+</output>

AGENTS.md

@@ -0,0 +1,58 @@
+# context-mode — MANDATORY routing rules
+
+You have context-mode MCP tools available. These rules are NOT optional — they protect your context window from flooding. A single unrouted command can dump 56 KB into context and waste the entire session.
+
+## BLOCKED commands — do NOT attempt these
+
+### curl / wget — BLOCKED
+Any shell command containing `curl` or `wget` will be intercepted and blocked by the context-mode plugin. Do NOT retry.
+Instead use:
+- `context-mode_ctx_fetch_and_index(url, source)` to fetch and index web pages
+- `context-mode_ctx_execute(language: "javascript", code: "const r = await fetch(...)")` to run HTTP calls in sandbox
+
+### Inline HTTP — BLOCKED
+Any shell command containing `fetch('http`, `requests.get(`, `requests.post(`, `http.get(`, or `http.request(` will be intercepted and blocked. Do NOT retry with shell.
+Instead use:
+- `context-mode_ctx_execute(language, code)` to run HTTP calls in sandbox — only stdout enters context
+
+### Direct web fetching — BLOCKED
+Do NOT use any direct URL fetching tool. Use the sandbox equivalent.
+Instead use:
+- `context-mode_ctx_fetch_and_index(url, source)` then `context-mode_ctx_search(queries)` to query the indexed content
+
+## REDIRECTED tools — use sandbox equivalents
+
+### Shell (>20 lines output)
+Shell is ONLY for: `git`, `mkdir`, `rm`, `mv`, `cd`, `ls`, `npm install`, `pip install`, and other short-output commands.
+For everything else, use:
+- `context-mode_ctx_batch_execute(commands, queries)` — run multiple commands + search in ONE call
+- `context-mode_ctx_execute(language: "shell", code: "...")` — run in sandbox, only stdout enters context
+
+### File reading (for analysis)
+If you are reading a file to **edit** it → reading is correct (edit needs content in context).
+If you are reading to **analyze, explore, or summarize** → use `context-mode_ctx_execute_file(path, language, code)` instead. Only your printed summary enters context.
+
+### grep / search (large results)
+Search results can flood context. Use `context-mode_ctx_execute(language: "shell", code: "grep ...")` to run searches in sandbox. Only your printed summary enters context.
+
+## Tool selection hierarchy
+
+1. **GATHER**: `context-mode_ctx_batch_execute(commands, queries)` — Primary tool. Runs all commands, auto-indexes output, returns search results. ONE call replaces 30+ individual calls.
+2. **FOLLOW-UP**: `context-mode_ctx_search(queries: ["q1", "q2", ...])` — Query indexed content. Pass ALL questions as array in ONE call.
+3. **PROCESSING**: `context-mode_ctx_execute(language, code)` | `context-mode_ctx_execute_file(path, language, code)` — Sandbox execution. Only stdout enters context.
+4. **WEB**: `context-mode_ctx_fetch_and_index(url, source)` then `context-mode_ctx_search(queries)` — Fetch, chunk, index, query. Raw HTML never enters context.
+5. **INDEX**: `context-mode_ctx_index(content, source)` — Store content in FTS5 knowledge base for later search.
+
+## Output constraints
+
+- Keep responses under 500 words.
+- Write artifacts (code, configs, PRDs) to FILES — never return them as inline text. Return only: file path + 1-line description.
+- When indexing content, use descriptive source labels so others can `search(source: "label")` later.
+
+## ctx commands
+
+| Command | Action |
+|---------|--------|
+| `ctx stats` | Call the `stats` MCP tool and display the full output verbatim |
+| `ctx doctor` | Call the `doctor` MCP tool, run the returned shell command, display as checklist |
+| `ctx upgrade` | Call the `upgrade` MCP tool, run the returned shell command, display as checklist |

Cargo.lock

@@ -92,6 +92,15 @@ dependencies = [
  "uuid",
 ]
 
+[[package]]
+name = "arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3d036a3c4ab069c7b410a2ce876bd74808d2d0888a82667669f8e783a898bf1"
+dependencies = [
+ "derive_arbitrary",
+]
+
 [[package]]
 name = "async-stream"
 version = "0.3.6"
@@ -202,6 +211,15 @@ version = "2.11.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "843867be96c8daad0d758b57df9392b6d8d271134fce549de6ce169ff98a92af"
 
+[[package]]
+name = "block-buffer"
+version = "0.10.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3078c7629b62d3f0439517fa394996acacc5cbc91c5a20d8c658e77abd503a71"
+dependencies = [
+ "generic-array",
+]
+
 [[package]]
 name = "bumpalo"
 version = "3.20.2"
@@ -296,6 +314,15 @@ version = "0.8.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "773648b94d0e5d620f64f280777445740e61fe701025087ec8b57f45c791888b"
 
+[[package]]
+name = "cpufeatures"
+version = "0.2.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "59ed5838eebb26a2bb2e58f6d5b5316989ae9d08bab10e0e6d103e656d1b0280"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "crc32fast"
 version = "1.5.0"
@@ -305,6 +332,64 @@ dependencies = [
  "cfg-if",
 ]
 
+[[package]]
+name = "crossbeam-utils"
+version = "0.8.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d0a5c400df2834b80a4c3327b3aad3a4c4cd4de0629063962b03235697506a28"
+
+[[package]]
+name = "crypto-common"
+version = "0.1.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "78c8292055d1c1df0cce5d180393dc8cce0abec0a7102adb6c7b1eef6016d60a"
+dependencies = [
+ "generic-array",
+ "typenum",
+]
+
+[[package]]
+name = "derive_arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e567bd82dcff979e4b03460c307b3cdc9e96fde3d73bed1496d2bc75d9dd62a"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "digest"
+version = "0.10.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9ed9a281f7bc9b7576e61468ba615a66a5c8cfdff42420a70aa82701a3b1e292"
+dependencies = [
+ "block-buffer",
+ "crypto-common",
+]
+
+[[package]]
+name = "dirs"
+version = "5.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "44c45a9d03d6676652bcb5e724c7e988de1acad23a711b5217ab9cbecbec2225"
+dependencies = [
+ "dirs-sys",
+]
+
+[[package]]
+name = "dirs-sys"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "520f05a5cbd335fae5a99ff7a6ab8627577660ee5cfd6a94a6a929b52ff0321c"
+dependencies = [
+ "libc",
+ "option-ext",
+ "redox_users",
+ "windows-sys 0.48.0",
+]
+
 [[package]]
 name = "displaydoc"
 version = "0.2.5"
@@ -457,6 +542,16 @@ dependencies = [
  "slab",
 ]
 
+[[package]]
+name = "generic-array"
+version = "0.14.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "85649ca51fd72272d7821adaf274ad91c288277713d9c18820d8499a7ff69e9a"
+dependencies = [
+ "typenum",
+ "version_check",
+]
+
 [[package]]
 name = "getrandom"
 version = "0.2.17"
@@ -758,7 +853,7 @@ dependencies = [
  "llama-sys",
  "serde",
  "serde_json",
- "thiserror",
+ "thiserror 2.0.18",
  "tokio",
  "tokio-stream",
  "tracing",
@@ -774,13 +869,15 @@ dependencies = [
  "async-stream",
  "axum",
  "bytes",
+ "dirs",
  "futures",
  "hyper",
  "inference-core",
  "metrics",
+ "runtime-deps",
  "serde",
  "serde_json",
- "thiserror",
+ "thiserror 2.0.18",
  "tokio",
  "tokio-stream",
  "tower",
@@ -840,6 +937,15 @@ dependencies = [
  "windows-link",
 ]
 
+[[package]]
+name = "libredox"
+version = "0.1.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7ddbf48fd451246b1f8c2610bd3b4ac0cc6e149d89832867093ab69a17194f08"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "litemap"
 version = "0.8.1"
@@ -971,6 +1077,12 @@ version = "1.70.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "384b8ab6d37215f3c5301a95a4accb5d64aa607f1fcb26a11b5303878451b4fe"
 
+[[package]]
+name = "option-ext"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "04744f49eae99ab78e0d5c0b603ab218f515ea8cfe5a456d7629ad883a3b6e7d"
+
 [[package]]
 name = "parking_lot"
 version = "0.12.5"
@@ -1064,6 +1176,17 @@ dependencies = [
  "bitflags",
 ]
 
+[[package]]
+name = "redox_users"
+version = "0.4.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba009ff324d1fc1b900bd1fdb31564febe58a8ccc8a6fdbb93b543d33b13ca43"
+dependencies = [
+ "getrandom 0.2.17",
+ "libredox",
+ "thiserror 1.0.69",
+]
+
 [[package]]
 name = "regex-automata"
 version = "0.4.14"
@@ -1095,6 +1218,19 @@ dependencies = [
  "windows-sys 0.52.0",
 ]
 
+[[package]]
+name = "runtime-deps"
+version = "0.1.0"
+dependencies = [
+ "anyhow",
+ "dirs",
+ "serde",
+ "serde_json",
+ "sha2",
+ "ureq",
+ "zip",
+]
+
 [[package]]
 name = "rustls"
 version = "0.23.37"
@@ -1220,6 +1356,17 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "sha2"
+version = "0.10.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a7507d819769d01a365ab707794a4084392c824f54a7a6a7862f8c3d0892b283"
+dependencies = [
+ "cfg-if",
+ "cpufeatures",
+ "digest",
+]
+
 [[package]]
 name = "sharded-slab"
 version = "0.1.7"
@@ -1319,13 +1466,33 @@ dependencies = [
  "syn",
 ]
 
+[[package]]
+name = "thiserror"
+version = "1.0.69"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b6aaf5339b578ea85b50e080feb250a3e8ae8cfcdff9a461c9ec2904bc923f52"
+dependencies = [
+ "thiserror-impl 1.0.69",
+]
+
 [[package]]
 name = "thiserror"
 version = "2.0.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4288b5bcbc7920c07a1149a35cf9590a2aa808e0bc1eafaade0b80947865fbc4"
 dependencies = [
- "thiserror-impl",
+ "thiserror-impl 2.0.18",
+]
+
+[[package]]
+name = "thiserror-impl"
+version = "1.0.69"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4fee6c4efc90059e10f81e6d42c60a18f76588c3d74cb83a0b242a2b6c7504c1"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
 ]
 
 [[package]]
@@ -1535,6 +1702,12 @@ version = "0.2.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e421abadd41a4225275504ea4d6566923418b7f05506fbc9c0fe86ba7396114b"
 
+[[package]]
+name = "typenum"
+version = "1.19.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "562d481066bde0658276a35467c4af00bdc6ee726305698a55b86e61d7ad82bb"
+
 [[package]]
 name = "unicode-ident"
 version = "1.0.24"
@@ -1612,6 +1785,12 @@ version = "0.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ba73ea9cf16a25df0c8caa16c51acb937d5712a8429db78a3ee29d5dcacd3a65"
 
+[[package]]
+name = "version_check"
+version = "0.9.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b928f33d975fc6ad9f86c8f283853ad26bdd5b10b7f1542aa2fa15e2289105a"
+
 [[package]]
 name = "want"
 version = "0.3.1"
@@ -1801,13 +1980,22 @@ dependencies = [
  "windows-link",
 ]
 
+[[package]]
+name = "windows-sys"
+version = "0.48.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "677d2418bec65e3338edb076e806bc1ec15693c5d0104683f2efe857f61056a9"
+dependencies = [
+ "windows-targets 0.48.5",
+]
+
 [[package]]
 name = "windows-sys"
 version = "0.52.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "282be5f36a8ce781fad8c8ae18fa3f9beff57ec1b52cb3de0789201425d9a33d"
 dependencies = [
- "windows-targets",
+ "windows-targets 0.52.6",
 ]
 
 [[package]]
@@ -1819,34 +2007,67 @@ dependencies = [
  "windows-link",
 ]
 
+[[package]]
+name = "windows-targets"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9a2fa6e2155d7247be68c096456083145c183cbbbc2764150dda45a87197940c"
+dependencies = [
+ "windows_aarch64_gnullvm 0.48.5",
+ "windows_aarch64_msvc 0.48.5",
+ "windows_i686_gnu 0.48.5",
+ "windows_i686_msvc 0.48.5",
+ "windows_x86_64_gnu 0.48.5",
+ "windows_x86_64_gnullvm 0.48.5",
+ "windows_x86_64_msvc 0.48.5",
+]
+
 [[package]]
 name = "windows-targets"
 version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9b724f72796e036ab90c1021d4780d4d3d648aca59e491e6b98e725b84e99973"
 dependencies = [
- "windows_aarch64_gnullvm",
- "windows_aarch64_msvc",
- "windows_i686_gnu",
+ "windows_aarch64_gnullvm 0.52.6",
+ "windows_aarch64_msvc 0.52.6",
+ "windows_i686_gnu 0.52.6",
  "windows_i686_gnullvm",
- "windows_i686_msvc",
- "windows_x86_64_gnu",
- "windows_x86_64_gnullvm",
- "windows_x86_64_msvc",
+ "windows_i686_msvc 0.52.6",
+ "windows_x86_64_gnu 0.52.6",
+ "windows_x86_64_gnullvm 0.52.6",
+ "windows_x86_64_msvc 0.52.6",
 ]
 
+[[package]]
+name = "windows_aarch64_gnullvm"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2b38e32f0abccf9987a4e3079dfb67dcd799fb61361e53e2882c3cbaf0d905d8"
+
 [[package]]
 name = "windows_aarch64_gnullvm"
 version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "32a4622180e7a0ec044bb555404c800bc9fd9ec262ec147edd5989ccd0c02cd3"
 
+[[package]]
+name = "windows_aarch64_msvc"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dc35310971f3b2dbbf3f0690a219f40e2d9afcf64f9ab7cc1be722937c26b4bc"
+
 [[package]]
 name = "windows_aarch64_msvc"
 version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "09ec2a7bb152e2252b53fa7803150007879548bc709c039df7627cabbd05d469"
 
+[[package]]
+name = "windows_i686_gnu"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a75915e7def60c94dcef72200b9a8e58e5091744960da64ec734a6c6e9b3743e"
+
 [[package]]
 name = "windows_i686_gnu"
 version = "0.52.6"
@@ -1859,24 +2080,48 @@ version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0eee52d38c090b3caa76c563b86c3a4bd71ef1a819287c19d586d7334ae8ed66"
 
+[[package]]
+name = "windows_i686_msvc"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8f55c233f70c4b27f66c523580f78f1004e8b5a8b659e05a4eb49d4166cca406"
+
 [[package]]
 name = "windows_i686_msvc"
 version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "240948bc05c5e7c6dabba28bf89d89ffce3e303022809e73deaefe4f6ec56c66"
 
+[[package]]
+name = "windows_x86_64_gnu"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "53d40abd2583d23e4718fddf1ebec84dbff8381c07cae67ff7768bbf19c6718e"
+
 [[package]]
 name = "windows_x86_64_gnu"
 version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "147a5c80aabfbf0c7d901cb5895d1de30ef2907eb21fbbab29ca94c5b08b1a78"
 
+[[package]]
+name = "windows_x86_64_gnullvm"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b7b52767868a23d5bab768e390dc5f5c55825b6d30b86c844ff2dc7414044cc"
+
 [[package]]
 name = "windows_x86_64_gnullvm"
 version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "24d5b23dc417412679681396f2b49f3de8c1473deb516bd34410872eff51ed0d"
 
+[[package]]
+name = "windows_x86_64_msvc"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ed94fce61571a4006852b7389a063ab983c02eb1bb37b47f8272ce92d06d9538"
+
 [[package]]
 name = "windows_x86_64_msvc"
 version = "0.52.6"
@@ -2060,8 +2305,37 @@ dependencies = [
  "syn",
 ]
 
+[[package]]
+name = "zip"
+version = "2.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fabe6324e908f85a1c52063ce7aa26b68dcb7eb6dbc83a2d148403c9bc3eba50"
+dependencies = [
+ "arbitrary",
+ "crc32fast",
+ "crossbeam-utils",
+ "displaydoc",
+ "flate2",
+ "indexmap",
+ "memchr",
+ "thiserror 2.0.18",
+ "zopfli",
+]
+
 [[package]]
 name = "zmij"
 version = "1.0.21"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b8848ee67ecc8aedbaf3e4122217aff892639231befc6a1b58d29fff4c2cabaa"
+
+[[package]]
+name = "zopfli"
+version = "0.8.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f05cd8797d63865425ff89b5c4a48804f35ba0ce8d125800027ad6017d2b5249"
+dependencies = [
+ "bumpalo",
+ "crc32fast",
+ "log",
+ "simd-adler32",
+]

Cargo.toml

@@ -2,6 +2,7 @@
 resolver = "2"
 members = [
     "crates/llama-sys",
+    "crates/runtime-deps",
     "crates/inference-core",
     "crates/inference-server",
     "crates/api-contracts",
@@ -47,6 +48,9 @@ tokio-stream = "0.1"
 bytes = "1"
 clap = { version = "4", features = ["derive"] }
 ureq = { version = "2", features = ["json"] }
+sha2 = "0.10"
+zip = { version = "2", default-features = false, features = ["deflate"] }
+dirs = "5"
 
 [profile.dev]
 opt-level = 0

README.MD

@@ -1,98 +0,0 @@
-# ForgeLM
-
-ForgeLM e um workspace Rust para inferencia local de LLMs com tres superficies principais:
-
-- uma API HTTP local compativel com partes da API de chat da OpenAI
-- uma CLI (`localllm`) para operacao, diagnostico e automacao
-- um app desktop Tauri para runtime management e chat persistente
-
-O projeto hoje e organizado para rodar localmente, com foco em modelos `.gguf`, operacao local-first e uma ponte entre backend Rust, CLI e interface desktop.
-
-## O que existe hoje
-
-- `crates/inference-server`: servidor HTTP com `GET /health`, `GET /v1/models` e `POST /v1/chat/completions`
-- `crates/localllm-cli`: CLI para status, modelos, auth, config, benchmark, logs e doctor
-- `tauri-app`: app desktop com runtime manager, chat UI, armazenamento local e integracao com o backend
-- `crates/api-contracts`: contratos de request/response compartilhados
-- `crates/inference-core`: engine, perfis de runtime e pipeline de inferencia
-
-## Inicio rapido
-
-### 1. Subir a API local
-
-Defina pelo menos o caminho do modelo e execute o servidor:
-
-```bash
-set LOCALLLM_MODEL=D:\caminho\modelo.gguf
-cargo run -p inference-server
-```
-
-Validacoes basicas:
-
-```bash
-curl http://127.0.0.1:8080/health
-curl http://127.0.0.1:8080/v1/models
-```
-
-### 2. Usar a CLI
-
-```bash
-cargo run -p localllm-cli -- status
-cargo run -p localllm-cli -- models list
-```
-
-### 3. Abrir o app desktop
-
-```bash
-cd tauri-app
-npm install
-npm run tauri dev
-```
-
-No app, abra o Runtime Manager, configure modo local ou remoto e depois use o chat.
-
-## Requisitos
-
-- Rust toolchain atual para o workspace
-- Node.js/NPM para o app Tauri
-- um modelo `.gguf` local para o runtime
-- GPU NVIDIA e opcional; o projeto tambem possui perfil CPU fallback
-
-## Estrutura do repositorio
-
-- `crates/api-contracts/` - contratos HTTP e tipos compartilhados
-- `crates/inference-core/` - engine, sessoes, sampler e perfis de runtime
-- `crates/inference-server/` - servidor HTTP local
-- `crates/localllm-cli/` - CLI de operacao
-- `crates/metrics/` - contadores e medicao de throughput
-- `crates/llama-sys/` - bindings/integracao de baixo nivel
-- `tauri-app/` - app desktop Tauri + frontend React
-- `docs/` - documentacao detalhada
-
-## Documentacao
-
-- `docs/README.md` - indice da documentacao
-- `docs/api/overview.md` - visao geral do backend e da API
-- `docs/api/endpoints.md` - referencia HTTP
-- `docs/api/configuration.md` - variaveis de ambiente e configuracao do servidor
-- `docs/cli/overview.md` - visao geral da CLI
-- `docs/cli/commands.md` - referencia de comandos e exemplos
-- `docs/app/overview.md` - visao geral do app desktop
-- `docs/app/user-guide.md` - guia de uso do app
-- `docs/app/backend-commands.md` - comandos Tauri expostos pelo backend do app
-- `docs/developer/architecture.md` - arquitetura do workspace e fluxos entre modulos
-
-## Estado atual e limitacoes
-
-- A API atual expoe apenas tres endpoints publicos: `GET /health`, `GET /v1/models` e `POST /v1/chat/completions`.
-- A compatibilidade com OpenAI e parcial: o projeto foca no fluxo de chat, nao em toda a superficie da plataforma.
-- A CLI possui uma superficie ampla, mas a execucao ao vivo hoje esta mais consolidada em `status`, `models list` e `bench throughput`.
-- O app desktop controla o runtime local no modo `local` e pode consumir uma API ja existente no modo `remote`.
-- O armazenamento do chat no frontend usa `sql.js` carregado por CDN e persistencia em `localStorage`.
-
-## Leitura recomendada
-
-- Quer entender a API? Comece em `docs/api/overview.md`.
-- Quer operar o sistema por terminal? Va para `docs/cli/overview.md`.
-- Quer usar o app desktop? Leia `docs/app/user-guide.md`.
-- Quer contribuir no codigo? Comece em `docs/developer/architecture.md`.

README.md

@@ -0,0 +1,123 @@
+<div align="center">
+
+<img src="assets/readme/hero.svg" width="100%" alt="ForgeLM: roda modelos .gguf na sua máquina com uma API no estilo OpenAI, em Rust sobre o llama.cpp">
+
+![Rust 2021](https://img.shields.io/badge/Rust-edi%C3%A7%C3%A3o_2021-dea584?style=flat-square&logo=rust&logoColor=white)
+![llama.cpp b8358](https://img.shields.io/badge/llama.cpp-b8358-4b86ff?style=flat-square)
+![Tauri 2](https://img.shields.io/badge/Tauri-2-24C8DB?style=flat-square&logo=tauri&logoColor=white)
+![Status](https://img.shields.io/badge/status-em_desenvolvimento-yellow?style=flat-square)
+
+[Início rápido](#in%C3%ADcio-r%C3%A1pido) · [Documentação](docs/README.md) · [Issues](https://git.evandro.dev.br/evandro/ForgeLM/issues)
+
+</div>
+
+<p align="center">
+  <img src="loaded-shell.png" width="720" alt="App desktop do ForgeLM: Runtime Manager com a API local parada, lista de conversas vazia e área de chat">
+</p>
+<p align="center"><sub>Captura do import inicial (março de 2026), com a API local parada. A interface mudou depois: o seletor de modelo foi para a caixa de mensagem.</sub></p>
+
+## O que é
+
+O ForgeLM roda modelos `.gguf` na sua máquina com o [llama.cpp](https://github.com/ggml-org/llama.cpp) e os expõe numa API HTTP no estilo OpenAI, com streaming SSE. O mesmo workspace Rust traz uma CLI (`localllm`) e um app desktop em Tauri que liga o servidor, busca e baixa modelos do [Hugging Face](https://huggingface.co) e conversa com eles.
+
+## Por que usar
+
+- **Uma peça só, local.** Servidor, engine, download do runtime e app vivem num workspace Rust. Na primeira execução o servidor baixa o llama.cpp `b8358` (CPU ou CUDA 12.4) do GitHub e confere o SHA-256, se não achar um runtime já instalado.
+- **Clientes OpenAI funcionam com poucas mudanças.** `GET /v1/models` e `POST /v1/chat/completions` seguem o formato da OpenAI; basta trocar a URL base.
+- **Perfis prontos.** `Turbo4B` (padrão), `balanced_9b` e `cpu_fallback` definem camadas na GPU, contexto e threads, e cada valor pode ser sobrescrito por variável de ambiente.
+
+**Quando não usar:** se você precisa de compatibilidade ampla com a OpenAI (são três endpoints, `n` precisa ser `1`, sem embeddings), de download automático do runtime fora do Windows x64, ou de algo estável. Nesses casos, o servidor do próprio llama.cpp atende melhor.
+
+## Como funciona
+
+- `inference-server` (binário `localllm`): API em Axum na porta `8080`. Valida autenticação, CORS e aliases e repassa ao `inference-core`.
+- `inference-core`: engine, sessão, sampler e perfis, sobre o `llama-sys`, que carrega o `llama.dll` em tempo de execução.
+- `runtime-deps`: acha ou baixa o runtime descrito em [`vendor/llama-runtime/manifest.json`](vendor/llama-runtime/manifest.json).
+- `localllm-cli`: cliente HTTP da mesma API, com `status`, `models`, `auth`, `config`, `bench`, `logs` e `doctor`.
+- `tauri-app`: React 18 + Vite + Tailwind com backend Tauri 2. No modo `local` inicia e para o servidor e consulta `GET /health`; no modo `remote` só fala com a API. A busca de modelos no Hugging Face não passa pela API local.
+
+Detalhes em [`docs/developer/architecture.md`](docs/developer/architecture.md).
+
+## Início rápido
+
+> [!WARNING]
+> **A API sobe aberta por padrão.** Sem `LOCALLLM_API_KEY` não há autenticação, o bind padrão é `0.0.0.0:8080` e o CORS libera `*`. Fora de uma máquina confiável, defina a chave e restrinja `LOCALLLM_ADDR` e `LOCALLLM_CORS_ORIGINS`.
+
+Requisitos: Rust (cargo; o projeto não declara `rust-version`), um modelo `.gguf` local e, só para o app, Node.js com npm (sem versão declarada). GPU NVIDIA é opcional: o perfil `cpu_fallback` roda sem ela.
+
+Na primeira execução, o download do runtime tem cerca de 37 MB na variante CPU e cerca de 618 MB na CUDA 12.4 (dois zips). O manifesto só lista `windows-x64`; em outros sistemas, aponte `LLAMA_DLL_PATH` para uma biblioteca já instalada.
+
+```bash
+# 1. API local (bash; no cmd do Windows, use set no lugar de export)
+export LOCALLLM_MODEL=/caminho/para/meu-modelo.gguf
+cargo run -p inference-server
+
+curl http://127.0.0.1:8080/health
+curl http://127.0.0.1:8080/v1/models
+curl -N http://127.0.0.1:8080/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{"model":"default","messages":[{"role":"user","content":"Diga olá"}],"stream":true}'
+
+# 2. CLI
+cargo run -p localllm-cli -- status
+cargo run -p localllm-cli -- models list
+
+# 3. App desktop
+cd tauri-app
+npm install
+npm run tauri dev
+```
+
+Testes: `cargo test` (crates do workspace), `cargo test --manifest-path tauri-app/src-tauri/Cargo.toml` (backend Tauri, fora do workspace) e `npm test` em `tauri-app` (Vitest).
+
+## Configuração
+
+A API lê tudo de variáveis de ambiente; não há `.env.example`. A pasta de modelos não é variável: na CLI é a flag global `--models-dir` (padrão `./models`), no app é o campo `models_dir` da configuração do runtime. Ao iniciar o servidor local, o app define `LOCALLLM_MODEL`, `LOCALLLM_ADDR`, `LOCALLLM_GPU_LAYERS` e `LOCALLLM_CTX` (contexto padrão do app: 4096).
+
+<details>
+<summary>Variáveis de ambiente da API</summary>
+
+| Variável | Obrigatória | Descrição |
+|---|---|---|
+| `LOCALLLM_MODEL` | Sim | Caminho do `.gguf`; sem ela o engine aborta (`model_path is empty`) |
+| `LOCALLLM_ADDR` | Não | Endereço de bind; padrão `0.0.0.0:8080` |
+| `LOCALLLM_MODEL_ALIASES` | Não | Aliases do modelo, separados por vírgula, expostos em `GET /v1/models` |
+| `LOCALLLM_PROFILE` | Não | `balanced_9b` ou `cpu_fallback`; ausente ou outro valor = `Turbo4B` |
+| `LOCALLLM_GPU_LAYERS` | Não | Camadas na GPU; padrão do perfil (999 em `Turbo4B` e `balanced_9b`, 0 em `cpu_fallback`) |
+| `LOCALLLM_CTX` | Não | Janela de contexto; padrão do perfil (131072, 65536 e 16384) |
+| `LOCALLLM_BATCH` | Não | `n_batch`; sem valor explícito, usa o do perfil limitado a 4096 |
+| `LOCALLLM_UBATCH` | Não | `n_ubatch`; padrão do perfil |
+| `LOCALLLM_THREADS` | Não | Threads de geração; padrão do perfil |
+| `LOCALLLM_THREADS_BATCH` | Não | Threads de prefill; padrão do perfil |
+| `LOCALLLM_API_KEY` | Não | Liga a autenticação por chave; vazia ou ausente = API aberta |
+| `LOCALLLM_AUTH_HEADER` | Não | Cabeçalho preferencial da chave; padrão `authorization` (também aceita `x-api-key`) |
+| `LOCALLLM_AUTH_SCHEME` | Não | Prefixo removido antes da comparação (ex.: `Bearer`) |
+| `LOCALLLM_CORS_ORIGINS` | Não | Origens permitidas, separadas por vírgula; padrão `*` |
+| `LOCALLLM_SYSTEM_PROMPT` | Não | Prompt de sistema usado quando a request não traz `system` nem `developer` |
+| `LOCALLLM_ENABLE_TOOL_CALLS` | Não | Liga ou desliga tool calling; padrão ligado |
+| `LOCALLLM_LOG_FORMAT` | Não | `json` para logs estruturados; padrão `pretty` |
+| `LLAMA_DLL_PATH` | Não | Caminho completo da biblioteca do llama.cpp; sobrepõe a resolução automática |
+| `RUST_LOG` | Não | Filtro de logs do `tracing`; padrão `info` |
+
+</details>
+
+## Estado do projeto
+
+Em desenvolvimento. Pelo [`.planning/ROADMAP.md`](.planning/ROADMAP.md), as fases 1 a 4 estão concluídas, a fase 5 (validação de fontes de modelo) segue aberta com os três planos marcados e a fase 6 (redesign do chat e das Configurações) tem 0 de 3 planos.
+
+- A compatibilidade com a OpenAI é parcial: `parallel_tool_calls` não é recurso suportado, `content` em partes vira texto e o raciocínio não sai num campo separado (quem separa `<think>…</think>` é o app).
+- O chat do app usa `sql.js` carregado por CDN e guarda o banco no `localStorage`.
+- `scripts/build-dev.sh` e `scripts/build-prod.sh` (e as versões `.ps1`) exigem `vendor/llama.cpp`, que não está no repositório.
+- `inference-server` e `localllm-cli` declaram um binário com o mesmo nome, `localllm`; use `cargo run -p <crate>`.
+
+Para contribuir, abra uma issue ou um PR e rode os testes da parte que você alterou.
+
+## Licença
+
+Sem licença publicada: todos os direitos reservados. O [`Cargo.toml`](Cargo.toml) declara `license = "MIT"`, mas o repositório não inclui o arquivo `LICENSE`. Os binários do llama.cpp e os modelos do Hugging Face seguem a licença de cada autor.
+
+<div align="center">
+
+<sub>Feito por <a href="https://evandro.dev.br">Evandro Fonseca Junior</a> · <a href="https://git.evandro.dev.br/evandro">git.evandro.dev.br</a></sub>
+
+</div>

assets/readme/hero.svg

@@ -0,0 +1,60 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="1200" height="420" viewBox="0 0 1200 420" role="img" aria-labelledby="title desc">
+  <title id="title">ForgeLM</title>
+  <desc id="desc">ForgeLM: um arquivo .gguf entra no llama.cpp, que divide as camadas entre GPU e CPU, e a resposta sai em streaming pela API local na porta 8080, em /v1/chat/completions.</desc>
+
+  <rect width="1200" height="420" rx="24" fill="#171717"/>
+
+  <g id="title-block" font-family="system-ui, -apple-system, 'Segoe UI', Roboto, Helvetica, Arial, sans-serif">
+    <text x="64" y="170" font-size="64" font-weight="700" fill="#F3F4F6">ForgeLM</text>
+    <text x="64" y="226" font-size="28" fill="#F3F4F6">Rode modelos .gguf na sua máquina</text>
+    <text x="64" y="264" font-size="28" fill="#F3F4F6">com uma API no estilo OpenAI.</text>
+    <text x="64" y="322" font-size="20" fill="#9CA3AF" font-family="ui-monospace, 'SF Mono', Consolas, monospace">Rust · llama.cpp · Axum · Tauri 2</text>
+  </g>
+
+  <g id="model-file" font-family="ui-monospace, 'SF Mono', Consolas, monospace">
+    <path d="M584 152 h60 l28 28 v88 h-88 z" fill="#262626" stroke="#313944" stroke-width="2"/>
+    <path d="M644 152 v28 h28" fill="none" stroke="#313944" stroke-width="2"/>
+    <text x="628" y="234" font-size="20" fill="#F3F4F6" text-anchor="middle">.gguf</text>
+    <text x="628" y="304" font-size="18" fill="#9CA3AF" text-anchor="middle" font-family="system-ui, -apple-system, 'Segoe UI', Roboto, Helvetica, Arial, sans-serif">modelo</text>
+  </g>
+
+  <g id="arrows" fill="none" stroke="#9CA3AF" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <path d="M682 210 h26 M700 202 l8 8 l-8 8"/>
+    <path d="M836 210 h20 M848 202 l8 8 l-8 8"/>
+  </g>
+
+  <g id="layers" font-family="system-ui, -apple-system, 'Segoe UI', Roboto, Helvetica, Arial, sans-serif">
+    <text x="772" y="110" font-size="20" font-weight="700" fill="#F3F4F6" text-anchor="middle">llama.cpp</text>
+    <rect x="716" y="126" width="112" height="28" rx="6" fill="#4b86ff"/>
+    <rect x="716" y="162" width="112" height="28" rx="6" fill="#4b86ff"/>
+    <rect x="716" y="198" width="112" height="28" rx="6" fill="#4b86ff"/>
+    <rect x="716" y="234" width="112" height="28" rx="6" fill="#4b86ff"/>
+    <rect x="716" y="270" width="112" height="28" rx="6" fill="#333333"/>
+    <rect x="716" y="306" width="112" height="28" rx="6" fill="#333333"/>
+    <text x="772" y="146" font-size="18" font-weight="700" fill="#0d0d10" text-anchor="middle">GPU</text>
+    <text x="772" y="326" font-size="18" font-weight="700" fill="#F3F4F6" text-anchor="middle">CPU</text>
+  </g>
+
+  <g id="api-stream" font-family="ui-monospace, 'SF Mono', Consolas, monospace">
+    <rect x="864" y="96" width="272" height="244" rx="14" fill="#0d0d10" stroke="#313944" stroke-width="2"/>
+    <circle cx="890" cy="128" r="6" fill="#30d158"/>
+    <text x="906" y="135" font-size="20" fill="#F3F4F6">:8080</text>
+    <text x="1116" y="135" font-size="18" fill="#9CA3AF" text-anchor="end">SSE</text>
+    <path d="M864 154 h272" stroke="#232a33" stroke-width="2"/>
+    <text x="884" y="186" font-size="18" fill="#9CA3AF">/v1/chat/completions</text>
+    <g font-family="system-ui, -apple-system, 'Segoe UI', Roboto, Helvetica, Arial, sans-serif" font-size="20" fill="#F3F4F6">
+      <rect x="884" y="204" width="58" height="34" rx="8" fill="#101c2d" stroke="#203451" stroke-width="2"/>
+      <text x="913" y="228" text-anchor="middle">Olá</text>
+      <rect x="950" y="204" width="30" height="34" rx="8" fill="#101c2d" stroke="#203451" stroke-width="2"/>
+      <text x="965" y="228" text-anchor="middle">,</text>
+      <rect x="988" y="204" width="68" height="34" rx="8" fill="#101c2d" stroke="#203451" stroke-width="2"/>
+      <text x="1022" y="228" text-anchor="middle">tudo</text>
+      <rect x="884" y="248" width="62" height="34" rx="8" fill="#101c2d" stroke="#203451" stroke-width="2"/>
+      <text x="915" y="272" text-anchor="middle">bem</text>
+      <rect x="954" y="248" width="30" height="34" rx="8" fill="#101c2d" stroke="#203451" stroke-width="2"/>
+      <text x="969" y="272" text-anchor="middle">?</text>
+      <rect x="994" y="252" width="12" height="26" rx="2" fill="#4b86ff"/>
+    </g>
+    <text x="884" y="318" font-size="18" fill="#9CA3AF">"stream": true</text>
+  </g>
+</svg>

crates/inference-core/src/profile.rs

@@ -11,29 +11,34 @@ pub enum RuntimeProfile {
 
 impl RuntimeProfile {
     /// Returns sensible defaults for each profile.
+    ///
+    /// `n_ctx` sets the KV cache / context window size. Larger values allow longer
+    /// conversations but consume more VRAM. Override at runtime with `LOCALLLM_CTX`.
+    /// `n_batch` controls the prefill chunk size — kept independent of `n_ctx` to
+    /// avoid pre-allocating the full window as a single batch buffer (OOM risk).
     pub fn defaults(&self) -> ProfileDefaults {
         match self {
             Self::Turbo4B => ProfileDefaults {
                 n_gpu_layers: 999,
-                n_ctx: 4096,
-                n_batch: 512,
-                n_ubatch: 128,
+                n_ctx: 131_072, // 128 K — covers very long conversations; needs ~4 GB KV VRAM at FP16
+                n_batch: 2048,  // prefill chunk size; independent of n_ctx
+                n_ubatch: 512,
                 flash_attn: true,
                 n_threads: 8,
             },
             Self::Balanced9B => ProfileDefaults {
                 n_gpu_layers: 999,
-                n_ctx: 2048,
-                n_batch: 256,
-                n_ubatch: 64,
+                n_ctx: 65_536, // 64 K
+                n_batch: 2048,
+                n_ubatch: 256,
                 flash_attn: true,
                 n_threads: 8,
             },
             Self::CpuFallback => ProfileDefaults {
                 n_gpu_layers: 0,
-                n_ctx: 2048,
-                n_batch: 128,
-                n_ubatch: 32,
+                n_ctx: 16_384, // 16 K — CPU KV cache is RAM-backed; keep conservative
+                n_batch: 512,
+                n_ubatch: 128,
                 flash_attn: false,
                 n_threads: num_cpus(),
             },

crates/inference-server/Cargo.toml

@@ -12,6 +12,8 @@ path = "src/main.rs"
 inference-core = { path = "../inference-core" }
 api-contracts = { path = "../api-contracts" }
 metrics = { path = "../metrics" }
+runtime-deps = { path = "../runtime-deps" }
+dirs = { workspace = true }
 tokio.workspace = true
 axum.workspace = true
 hyper.workspace = true

crates/inference-server/build.rs

@@ -1,38 +1,4 @@
 fn main() {
-    let out_dir = std::env::var("OUT_DIR").unwrap();
-    let target_dir = std::path::Path::new(&out_dir)
-        .ancestors()
-        .nth(3) // Volta 3 níveis de `target/debug/build/inference-server-XYZ/out` para `target/debug/` ou `target/release/`
-        .unwrap();
-
-    let vendor_dir = std::path::PathBuf::from(std::env::var("CARGO_MANIFEST_DIR").unwrap())
-        .parent()
-        .unwrap()
-        .parent()
-        .unwrap()
-        .join("vendor")
-        .join("llama-bin");
-
-    if vendor_dir.exists() {
-        println!(
-            "cargo:warning=Copying DLLs from {:?} to {:?}",
-            vendor_dir, target_dir
-        );
-
-        // Vamos copiar todos os arquivos .dll
-        if let Ok(entries) = std::fs::read_dir(&vendor_dir) {
-            for entry in entries.flatten() {
-                let path = entry.path();
-                if path.is_file() && path.extension().and_then(|s| s.to_str()) == Some("dll") {
-                    let file_name = path.file_name().unwrap();
-                    let dest = target_dir.join(file_name);
-                    let _ = std::fs::copy(&path, &dest);
-                }
-            }
-        }
-    }
-
-    // Rebuildar apenas se houver mudanças nessa build file ou nos bins da vendor
     println!("cargo:rerun-if-changed=build.rs");
-    println!("cargo:rerun-if-changed=../../vendor/llama-bin");
+    println!("cargo:rerun-if-changed=../../vendor/llama-runtime/manifest.json");
 }

crates/inference-server/src/main.rs

@@ -3,8 +3,12 @@ use std::path::PathBuf;
 use std::sync::Arc;
 
 use anyhow::Result;
+use runtime_deps::{
+    apply_runtime_to_current_process, ensure_llama_runtime, recommended_variant_override_hint,
+    EnsureRuntimeRequest,
+};
 use tokio::sync::Mutex;
-use tracing_subscriber::{fmt, EnvFilter};
+use tracing_subscriber::{fmt, layer::SubscriberExt, util::SubscriberInitExt, EnvFilter};
 
 use inference_core::{EngineConfig, InferenceEngine, RuntimeProfile};
 use metrics::counters::PerfCounters;
@@ -17,12 +21,25 @@ use routes::{build_router, AppState};
 #[tokio::main]
 async fn main() -> Result<()> {
     // ── Tracing ──────────────────────────────────────────────────────────
-    fmt()
-        .with_env_filter(
-            EnvFilter::try_from_default_env().unwrap_or_else(|_| EnvFilter::new("info")),
-        )
-        .with_target(false)
-        .init();
+    // Set LOCALLLM_LOG_FORMAT=json for machine-readable output (e.g. log aggregators).
+    // Set RUST_LOG=debug for verbose output.
+    let env_filter = EnvFilter::try_from_default_env()
+        .unwrap_or_else(|_| EnvFilter::new("info"));
+
+    match std::env::var("LOCALLLM_LOG_FORMAT").as_deref() {
+        Ok("json") => {
+            tracing_subscriber::registry()
+                .with(env_filter)
+                .with(fmt::layer().json().with_current_span(true).with_span_list(true))
+                .init();
+        }
+        _ => {
+            tracing_subscriber::registry()
+                .with(env_filter)
+                .with(fmt::layer().with_target(true))
+                .init();
+        }
+    }
 
     // ── Engine config (reads env vars; sane defaults for dev) ────────────
     let model_path = std::env::var("LOCALLLM_MODEL")
@@ -62,6 +79,45 @@ async fn main() -> Result<()> {
     let server_config = Arc::new(ServerConfig::from_env(model_name.clone()));
 
     // ── Load engine ──────────────────────────────────────────────────────
+    tracing::info!(
+        model = %model_path.display(),
+        profile = ?profile,
+        n_gpu_layers = config.n_gpu_layers,
+        n_ctx = config.n_ctx,
+        n_batch = config.n_batch,
+        n_ubatch = config.n_ubatch,
+        n_threads = config.n_threads,
+        n_threads_batch = config.n_threads_batch,
+        flash_attn = config.flash_attn,
+        use_mmap = config.use_mmap,
+        "engine_config"
+    );
+    let binary_path = std::env::current_exe()?;
+    let runtime_cache_root = dirs::data_local_dir()
+        .unwrap_or_else(|| {
+            binary_path
+                .parent()
+                .unwrap_or_else(|| std::path::Path::new("."))
+                .to_path_buf()
+        })
+        .join("ForgeLM");
+    let runtime = ensure_llama_runtime(&EnsureRuntimeRequest {
+        cache_root: &runtime_cache_root,
+        gpu_layers: config.n_gpu_layers,
+        variant_override: None,
+        binary_path: Some(&binary_path),
+    })
+    .map_err(|error| anyhow::anyhow!(
+        "failed to prepare llama runtime dependencies: {error}. {}",
+        recommended_variant_override_hint()
+    ))?;
+    apply_runtime_to_current_process(&runtime);
+    tracing::info!(
+        dll = %runtime.dll_path.display(),
+        variant = runtime.variant_id,
+        source = ?runtime.source,
+        "llama_runtime_ready"
+    );
     let mut engine = InferenceEngine::load(config)?;
     engine.warmup()?;
 
@@ -81,6 +137,14 @@ async fn main() -> Result<()> {
     let router = build_router(state);
     let listener = tokio::net::TcpListener::bind(addr).await?;
 
+    tracing::info!(
+        bind_addr = %addr,
+        model = %model_name,
+        api_key_set = std::env::var("LOCALLLM_API_KEY").is_ok(),
+        log_format = std::env::var("LOCALLLM_LOG_FORMAT").as_deref().unwrap_or("pretty"),
+        "server_ready"
+    );
+
     // ── Startup banner ───────────────────────────────────────────────────
     let display_addr = if addr.ip().is_unspecified() {
         format!("http://127.0.0.1:{}", addr.port())
@@ -125,8 +189,20 @@ fn env_u32(key: &str, default: u32) -> u32 {
         .unwrap_or(default)
 }
 
-fn resolve_batch_size(n_ctx: u32, default_n_batch: u32, explicit_batch: Option<u32>) -> u32 {
-    explicit_batch.unwrap_or_else(|| n_ctx.max(default_n_batch).max(1))
+/// Choose the prefill batch size.
+///
+/// The batch size controls how many tokens are processed in a single `llama_decode`
+/// call during prefill. It must not be confused with the context window (`n_ctx`):
+/// a large `n_ctx` does NOT require an equally large `n_batch` — `decode_prefill`
+/// already chunks the prompt into `n_batch`-sized slices.
+///
+/// Setting `n_batch = n_ctx` for a 128 K context would pre-allocate ~128 K token
+/// slots in the batch buffer, consuming gigabytes of VRAM at startup before any
+/// request arrives. Instead we cap auto-selection at `MAX_AUTO_BATCH` and let
+/// callers pass `LOCALLLM_BATCH` when they need a larger explicit override.
+fn resolve_batch_size(_n_ctx: u32, default_n_batch: u32, explicit_batch: Option<u32>) -> u32 {
+    const MAX_AUTO_BATCH: u32 = 4096;
+    explicit_batch.unwrap_or_else(|| default_n_batch.min(MAX_AUTO_BATCH).max(1))
 }
 
 #[cfg(test)]
@@ -134,8 +210,11 @@ mod tests {
     use super::resolve_batch_size;
 
     #[test]
-    fn defaults_batch_to_context_size_when_ctx_is_larger() {
-        assert_eq!(resolve_batch_size(16_384, 512, None), 16_384);
+    fn caps_auto_batch_at_max_regardless_of_ctx_size() {
+        // Large n_ctx must NOT bloat n_batch — that would OOM the batch buffer.
+        assert_eq!(resolve_batch_size(131_072, 512, None), 512);
+        assert_eq!(resolve_batch_size(131_072, 4096, None), 4096);
+        assert_eq!(resolve_batch_size(131_072, 8192, None), 4096); // capped at MAX_AUTO_BATCH
     }
 
     #[test]

crates/inference-server/src/routes.rs

@@ -397,9 +397,17 @@ async fn stream_response(
         }
     };
 
-    Sse::new(stream)
+    let response = Sse::new(stream)
         .keep_alive(KeepAlive::default())
-        .into_response()
+        .into_response();
+
+    axum::response::Response::builder()
+        .status(response.status())
+        .header("content-type", "text/event-stream; charset=utf-8")
+        .header("cache-control", "no-cache")
+        .header("connection", "keep-alive")
+        .body(response.into_body())
+        .unwrap()
 }
 
 async fn blocking_response(

crates/runtime-deps/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "runtime-deps"
+version.workspace = true
+edition.workspace = true
+
+[dependencies]
+anyhow = { workspace = true }
+serde = { workspace = true }
+serde_json = { workspace = true }
+sha2 = { workspace = true }
+ureq = { workspace = true }
+zip = { workspace = true }
+dirs = { workspace = true }

crates/runtime-deps/src/lib.rs

@@ -0,0 +1,614 @@
+use std::ffi::OsString;
+use std::fs::{self, File, OpenOptions};
+use std::io::{self, BufReader, Read, Write};
+use std::path::{Path, PathBuf};
+use std::process::Command;
+use std::thread;
+use std::time::{Duration, Instant};
+
+use anyhow::{anyhow, bail, Context, Result};
+use serde::Deserialize;
+use sha2::{Digest, Sha256};
+
+const MANIFEST_JSON: &str = include_str!("../../../vendor/llama-runtime/manifest.json");
+const LOCKFILE_NAME: &str = ".install.lock";
+const DOWNLOAD_TIMEOUT: Duration = Duration::from_secs(300);
+const LOCK_TIMEOUT: Duration = Duration::from_secs(300);
+const LOCK_POLL_INTERVAL: Duration = Duration::from_millis(250);
+
+#[derive(Debug, Clone)]
+pub struct EnsureRuntimeRequest<'a> {
+    pub cache_root: &'a Path,
+    pub gpu_layers: i32,
+    pub variant_override: Option<&'a str>,
+    pub binary_path: Option<&'a Path>,
+}
+
+#[derive(Debug, Clone)]
+pub struct ResolvedRuntime {
+    pub dll_path: PathBuf,
+    pub library_dir: PathBuf,
+    pub variant_id: String,
+    pub source: RuntimeSource,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum RuntimeSource {
+    EnvOverride,
+    AdjacentBinary,
+    LegacyVendor,
+    ManagedCache,
+}
+
+pub fn ensure_llama_runtime(request: &EnsureRuntimeRequest<'_>) -> Result<ResolvedRuntime> {
+    if let Some(runtime) = resolve_from_env_override()? {
+        return Ok(runtime);
+    }
+
+    let manifest = load_manifest()?;
+    let variant = select_variant(&manifest, request.gpu_layers, request.variant_override)?;
+
+    if let Some(runtime) = resolve_from_adjacent_binary(request.binary_path, variant)? {
+        return Ok(runtime);
+    }
+
+    if let Some(runtime) = resolve_from_legacy_vendor(request.binary_path, variant)? {
+        return Ok(runtime);
+    }
+    let install_dir = request
+        .cache_root
+        .join("llama-runtime")
+        .join(&manifest.release)
+        .join(&variant.id);
+    let dll_path = install_dir.join(&variant.dll);
+
+    if installation_is_valid(&install_dir, variant) {
+        return Ok(ResolvedRuntime {
+            dll_path,
+            library_dir: install_dir,
+            variant_id: variant.id.clone(),
+            source: RuntimeSource::ManagedCache,
+        });
+    }
+
+    fs::create_dir_all(install_dir.parent().unwrap_or(request.cache_root)).with_context(|| {
+        format!(
+            "failed to create runtime cache root at {}",
+            request.cache_root.display()
+        )
+    })?;
+    let _lock = InstallationLock::acquire(&install_dir)?;
+
+    if installation_is_valid(&install_dir, variant) {
+        return Ok(ResolvedRuntime {
+            dll_path,
+            library_dir: install_dir,
+            variant_id: variant.id.clone(),
+            source: RuntimeSource::ManagedCache,
+        });
+    }
+
+    install_variant(&install_dir, variant)?;
+
+    Ok(ResolvedRuntime {
+        dll_path,
+        library_dir: install_dir,
+        variant_id: variant.id.clone(),
+        source: RuntimeSource::ManagedCache,
+    })
+}
+
+pub fn apply_runtime_to_command(command: &mut Command, runtime: &ResolvedRuntime) {
+    command.env("LLAMA_DLL_PATH", &runtime.dll_path);
+    #[cfg(target_os = "windows")]
+    {
+        let path = prepend_path_entry(&runtime.library_dir, std::env::var_os("PATH"));
+        command.env("PATH", path);
+    }
+}
+
+pub fn apply_runtime_to_current_process(runtime: &ResolvedRuntime) {
+    std::env::set_var("LLAMA_DLL_PATH", &runtime.dll_path);
+    #[cfg(target_os = "windows")]
+    {
+        let path = prepend_path_entry(&runtime.library_dir, std::env::var_os("PATH"));
+        std::env::set_var("PATH", path);
+    }
+}
+
+pub fn recommended_variant_override_hint() -> &'static str {
+    "Set FORGELM_LLAMA_RUNTIME_VARIANT to a manifest variant id like windows-x64-cpu or windows-x64-cuda-12.4."
+}
+
+fn resolve_from_env_override() -> Result<Option<ResolvedRuntime>> {
+    let Ok(raw_path) = std::env::var("LLAMA_DLL_PATH") else {
+        return Ok(None);
+    };
+    let dll_path = PathBuf::from(raw_path);
+    if !dll_path.exists() {
+        bail!(
+            "LLAMA_DLL_PATH points to {}, but that file does not exist",
+            dll_path.display()
+        );
+    }
+    let library_dir = dll_path
+        .parent()
+        .map(Path::to_path_buf)
+        .ok_or_else(|| anyhow!("LLAMA_DLL_PATH must point to a file, not a root path"))?;
+    Ok(Some(ResolvedRuntime {
+        dll_path,
+        library_dir,
+        variant_id: "env-override".to_string(),
+        source: RuntimeSource::EnvOverride,
+    }))
+}
+
+fn resolve_from_adjacent_binary(
+    binary_path: Option<&Path>,
+    variant: &ManifestVariant,
+) -> Result<Option<ResolvedRuntime>> {
+    let Some(binary_path) = binary_path else {
+        return Ok(None);
+    };
+    let Some(binary_dir) = binary_path.parent() else {
+        return Ok(None);
+    };
+    let dll_path = binary_dir.join(default_dll_name());
+    if dll_path.exists() && verify_required_files(binary_dir, variant).is_ok() {
+        return Ok(Some(ResolvedRuntime {
+            dll_path,
+            library_dir: binary_dir.to_path_buf(),
+            variant_id: "adjacent-binary".to_string(),
+            source: RuntimeSource::AdjacentBinary,
+        }));
+    }
+    Ok(None)
+}
+
+fn resolve_from_legacy_vendor(
+    binary_path: Option<&Path>,
+    variant: &ManifestVariant,
+) -> Result<Option<ResolvedRuntime>> {
+    let Some(binary_path) = binary_path else {
+        return Ok(None);
+    };
+    let mut dir = binary_path.parent();
+    while let Some(current) = dir {
+        let candidate = current
+            .join("vendor")
+            .join("llama-bin")
+            .join(default_dll_name());
+        let Some(library_dir) = candidate.parent() else {
+            dir = current.parent();
+            continue;
+        };
+        if candidate.exists() && verify_required_files(library_dir, variant).is_ok() {
+            return Ok(Some(ResolvedRuntime {
+                dll_path: candidate.clone(),
+                library_dir: library_dir.to_path_buf(),
+                variant_id: "legacy-vendor".to_string(),
+                source: RuntimeSource::LegacyVendor,
+            }));
+        }
+        dir = current.parent();
+    }
+    Ok(None)
+}
+
+fn load_manifest() -> Result<Manifest> {
+    serde_json::from_str(MANIFEST_JSON).context("failed to parse embedded llama runtime manifest")
+}
+
+fn select_variant<'a>(
+    manifest: &'a Manifest,
+    gpu_layers: i32,
+    variant_override: Option<&str>,
+) -> Result<&'a ManifestVariant> {
+    let env_override = std::env::var("FORGELM_LLAMA_RUNTIME_VARIANT").ok();
+    if let Some(override_id) = variant_override.or(env_override.as_deref()) {
+        return manifest
+            .variants
+            .iter()
+            .find(|variant| variant.id == override_id)
+            .ok_or_else(|| anyhow!("unknown runtime variant '{override_id}'"));
+    }
+
+    let target = current_target();
+    let want_gpu = gpu_layers > 0;
+    manifest
+        .variants
+        .iter()
+        .find(|variant| variant.target == target && variant.gpu == want_gpu)
+        .or_else(|| {
+            manifest
+                .variants
+                .iter()
+                .find(|variant| variant.target == target && !variant.gpu)
+        })
+        .ok_or_else(|| anyhow!("no runtime bundle configured for target {target}"))
+}
+
+fn install_variant(install_dir: &Path, variant: &ManifestVariant) -> Result<()> {
+    if let Some(parent) = install_dir.parent() {
+        fs::create_dir_all(parent).with_context(|| {
+            format!(
+                "failed to create runtime install parent at {}",
+                parent.display()
+            )
+        })?;
+    }
+
+    let staging_dir = install_dir.with_extension("staging");
+    if staging_dir.exists() {
+        fs::remove_dir_all(&staging_dir).with_context(|| {
+            format!(
+                "failed to clean previous staging dir {}",
+                staging_dir.display()
+            )
+        })?;
+    }
+    fs::create_dir_all(&staging_dir)
+        .with_context(|| format!("failed to create staging dir {}", staging_dir.display()))?;
+
+    for archive in &variant.archives {
+        let archive_path = download_archive_to_file(archive, &staging_dir)?;
+        extract_zip_file(&archive_path, &staging_dir)?;
+    }
+
+    verify_required_files(&staging_dir, variant)?;
+
+    if install_dir.exists() {
+        fs::remove_dir_all(install_dir).with_context(|| {
+            format!(
+                "failed to replace old runtime dir {}",
+                install_dir.display()
+            )
+        })?;
+    }
+    fs::rename(&staging_dir, install_dir).or_else(|rename_error| {
+        copy_dir_recursive(&staging_dir, install_dir)
+            .and_then(|_| {
+                fs::remove_dir_all(&staging_dir)
+                    .with_context(|| format!("failed to remove staging dir {}", staging_dir.display()))
+            })
+            .map_err(|copy_error| {
+                anyhow!(
+                    "failed to finalize runtime install via rename ({rename_error}) and copy fallback ({copy_error})"
+                )
+            })
+    })?;
+
+    Ok(())
+}
+
+fn download_archive_to_file(archive: &ManifestArchive, staging_dir: &Path) -> Result<PathBuf> {
+    let downloads_dir = staging_dir.join(".downloads");
+    fs::create_dir_all(&downloads_dir)
+        .with_context(|| format!("failed to create downloads dir {}", downloads_dir.display()))?;
+    let archive_name = archive
+        .url
+        .rsplit('/')
+        .next()
+        .filter(|name| !name.is_empty())
+        .ok_or_else(|| anyhow!("archive url has no file name: {}", archive.url))?;
+    let archive_path = downloads_dir.join(archive_name);
+
+    let response = ureq::get(&archive.url)
+        .timeout(DOWNLOAD_TIMEOUT)
+        .call()
+        .with_context(|| format!("failed to download {}", archive.url))?;
+    let mut reader = response.into_reader();
+    let mut file = File::create(&archive_path)
+        .with_context(|| format!("failed to create archive file {}", archive_path.display()))?;
+    let mut hasher = Sha256::new();
+    let mut total_bytes = 0u64;
+    let mut buffer = [0u8; 64 * 1024];
+
+    loop {
+        let read = reader
+            .read(&mut buffer)
+            .with_context(|| format!("failed to stream downloaded bytes from {}", archive.url))?;
+        if read == 0 {
+            break;
+        }
+        total_bytes += read as u64;
+        if total_bytes > archive.size {
+            bail!(
+                "downloaded archive {} exceeded declared size {} bytes",
+                archive.url,
+                archive.size
+            );
+        }
+        hasher.update(&buffer[..read]);
+        file.write_all(&buffer[..read]).with_context(|| {
+            format!(
+                "failed to write downloaded archive to {}",
+                archive_path.display()
+            )
+        })?;
+    }
+
+    if total_bytes != archive.size {
+        bail!(
+            "downloaded archive {} size mismatch: expected {} bytes, got {}",
+            archive.url,
+            archive.size,
+            total_bytes
+        );
+    }
+
+    let expected = &archive.sha256;
+    let actual = format!("{:x}", hasher.finalize());
+    if actual != *expected {
+        bail!(
+            "downloaded archive {} failed sha256 verification: expected {}, got {}",
+            archive.url,
+            expected,
+            actual
+        );
+    }
+    Ok(archive_path)
+}
+
+fn extract_zip_file(archive_path: &Path, dest_dir: &Path) -> Result<()> {
+    let file = File::open(archive_path)
+        .with_context(|| format!("failed to open archive {}", archive_path.display()))?;
+    let mut archive =
+        zip::ZipArchive::new(BufReader::new(file)).context("failed to open zip archive")?;
+    for index in 0..archive.len() {
+        let mut entry = archive
+            .by_index(index)
+            .context("failed to read zip entry")?;
+        let Some(relative_path) = entry.enclosed_name().map(|path| path.to_path_buf()) else {
+            continue;
+        };
+        let output_path = dest_dir.join(relative_path);
+        if entry.is_dir() {
+            fs::create_dir_all(&output_path).with_context(|| {
+                format!("failed to create extracted dir {}", output_path.display())
+            })?;
+            continue;
+        }
+
+        if let Some(parent) = output_path.parent() {
+            fs::create_dir_all(parent)
+                .with_context(|| format!("failed to create parent dir {}", parent.display()))?;
+        }
+
+        let mut output = File::create(&output_path).with_context(|| {
+            format!("failed to create extracted file {}", output_path.display())
+        })?;
+        io::copy(&mut entry, &mut output)
+            .with_context(|| format!("failed to extract {}", output_path.display()))?;
+    }
+    Ok(())
+}
+
+fn verify_required_files(install_dir: &Path, variant: &ManifestVariant) -> Result<()> {
+    for required_file in &variant.required_files {
+        let candidate = install_dir.join(required_file);
+        if !candidate.exists() {
+            bail!(
+                "runtime bundle {} is missing required file {}",
+                variant.id,
+                candidate.display()
+            );
+        }
+    }
+    Ok(())
+}
+
+fn installation_is_valid(install_dir: &Path, variant: &ManifestVariant) -> bool {
+    install_dir.exists() && verify_required_files(install_dir, variant).is_ok()
+}
+
+fn default_dll_name() -> &'static str {
+    #[cfg(target_os = "windows")]
+    {
+        "llama.dll"
+    }
+    #[cfg(target_os = "linux")]
+    {
+        "libllama.so"
+    }
+    #[cfg(target_os = "macos")]
+    {
+        "libllama.dylib"
+    }
+}
+
+fn current_target() -> &'static str {
+    #[cfg(all(target_os = "windows", target_arch = "x86_64"))]
+    {
+        "windows-x86_64"
+    }
+    #[cfg(all(target_os = "windows", target_arch = "aarch64"))]
+    {
+        "windows-aarch64"
+    }
+    #[cfg(all(target_os = "linux", target_arch = "x86_64"))]
+    {
+        "linux-x86_64"
+    }
+    #[cfg(all(target_os = "macos", target_arch = "x86_64"))]
+    {
+        "macos-x86_64"
+    }
+    #[cfg(all(target_os = "macos", target_arch = "aarch64"))]
+    {
+        "macos-aarch64"
+    }
+}
+
+fn prepend_path_entry(entry: &Path, existing: Option<OsString>) -> OsString {
+    let mut paths = vec![entry.to_path_buf()];
+    if let Some(existing) = existing {
+        paths.extend(std::env::split_paths(&existing));
+    }
+    std::env::join_paths(paths).unwrap_or_else(|_| OsString::from(entry.as_os_str()))
+}
+
+fn copy_dir_recursive(source: &Path, destination: &Path) -> Result<()> {
+    fs::create_dir_all(destination)
+        .with_context(|| format!("failed to create {}", destination.display()))?;
+    for entry in
+        fs::read_dir(source).with_context(|| format!("failed to read {}", source.display()))?
+    {
+        let entry = entry?;
+        let source_path = entry.path();
+        let destination_path = destination.join(entry.file_name());
+        if source_path.is_dir() {
+            copy_dir_recursive(&source_path, &destination_path)?;
+        } else {
+            fs::copy(&source_path, &destination_path).with_context(|| {
+                format!(
+                    "failed to copy {} to {}",
+                    source_path.display(),
+                    destination_path.display()
+                )
+            })?;
+        }
+    }
+    Ok(())
+}
+
+struct InstallationLock {
+    path: PathBuf,
+    _file: File,
+}
+
+impl InstallationLock {
+    fn acquire(install_dir: &Path) -> Result<Self> {
+        fs::create_dir_all(install_dir.parent().unwrap_or(install_dir)).with_context(|| {
+            format!("failed to create lock parent for {}", install_dir.display())
+        })?;
+
+        let lock_path = install_dir.join(LOCKFILE_NAME);
+        let started = Instant::now();
+        loop {
+            match OpenOptions::new()
+                .create_new(true)
+                .write(true)
+                .open(&lock_path)
+            {
+                Ok(file) => {
+                    return Ok(Self {
+                        path: lock_path,
+                        _file: file,
+                    })
+                }
+                Err(error) if error.kind() == io::ErrorKind::AlreadyExists => {
+                    if lock_is_stale(&lock_path)? {
+                        let _ = fs::remove_file(&lock_path);
+                        continue;
+                    }
+                    if started.elapsed() >= LOCK_TIMEOUT {
+                        bail!(
+                            "timed out waiting for runtime install lock at {}",
+                            lock_path.display()
+                        );
+                    }
+                    thread::sleep(LOCK_POLL_INTERVAL);
+                }
+                Err(error) => {
+                    return Err(error).with_context(|| {
+                        format!(
+                            "failed to acquire runtime install lock at {}",
+                            lock_path.display()
+                        )
+                    })
+                }
+            }
+        }
+    }
+}
+
+impl Drop for InstallationLock {
+    fn drop(&mut self) {
+        let _ = fs::remove_file(&self.path);
+    }
+}
+
+fn lock_is_stale(lock_path: &Path) -> Result<bool> {
+    let metadata = match fs::metadata(lock_path) {
+        Ok(metadata) => metadata,
+        Err(error) if error.kind() == io::ErrorKind::NotFound => return Ok(false),
+        Err(error) => {
+            return Err(error).with_context(|| {
+                format!("failed to read lock metadata for {}", lock_path.display())
+            })
+        }
+    };
+    let modified = metadata
+        .modified()
+        .with_context(|| format!("failed to read lock timestamp for {}", lock_path.display()))?;
+    Ok(modified.elapsed().unwrap_or_default() > LOCK_TIMEOUT)
+}
+
+#[derive(Debug, Deserialize)]
+struct Manifest {
+    release: String,
+    variants: Vec<ManifestVariant>,
+}
+
+#[derive(Debug, Deserialize)]
+struct ManifestVariant {
+    id: String,
+    target: String,
+    gpu: bool,
+    dll: String,
+    archives: Vec<ManifestArchive>,
+    required_files: Vec<String>,
+}
+
+#[derive(Debug, Deserialize)]
+struct ManifestArchive {
+    url: String,
+    sha256: String,
+    size: u64,
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{current_target, load_manifest, prepend_path_entry, select_variant, Manifest};
+    use std::ffi::OsString;
+    use std::path::Path;
+
+    fn manifest() -> Manifest {
+        load_manifest().expect("manifest")
+    }
+
+    #[test]
+    fn selects_cpu_variant_for_local_cpu_mode() {
+        let manifest = manifest();
+        let variant = select_variant(&manifest, 0, None).expect("variant");
+        assert_eq!(variant.target, current_target());
+        assert!(!variant.gpu);
+    }
+
+    #[test]
+    fn selects_gpu_variant_when_gpu_layers_are_requested() {
+        let manifest = manifest();
+        let variant = select_variant(&manifest, 32, None).expect("variant");
+        assert_eq!(variant.target, current_target());
+        assert!(variant.gpu);
+    }
+
+    #[test]
+    fn accepts_explicit_variant_override() {
+        let manifest = manifest();
+        let variant = select_variant(&manifest, 0, Some("windows-x64-cpu")).expect("variant");
+        assert_eq!(variant.id, "windows-x64-cpu");
+    }
+
+    #[test]
+    fn prepends_runtime_dir_to_existing_path() {
+        let joined = prepend_path_entry(
+            Path::new("C:/runtime"),
+            Some(OsString::from("C:/windows;C:/tools")),
+        );
+        let rendered = joined.to_string_lossy();
+        assert!(rendered.starts_with("C:/runtime"));
+        assert!(rendered.contains("C:/windows"));
+    }
+}

tauri-app/package.json

@@ -11,10 +11,12 @@
     "tauri": "tauri"
   },
   "dependencies": {
+    "@phosphor-icons/react": "^2.1.10",
     "@tauri-apps/api": "^2",
     "@tauri-apps/plugin-shell": "^2",
     "dompurify": "^3.0.0",
     "highlight.js": "^11.0.0",
+    "lucide-react": "^1.7.0",
     "marked": "^12.0.0",
     "react": "^18.3.1",
     "react-dom": "^18.3.1"
@@ -25,7 +27,10 @@
     "@types/react": "^18.3.1",
     "@types/react-dom": "^18.3.1",
     "@vitejs/plugin-react": "^4.3.1",
+    "autoprefixer": "^10.4.27",
     "jsdom": "^29.0.1",
+    "postcss": "^8.5.8",
+    "tailwindcss": "^3.4.19",
     "typescript": "^5.5.3",
     "vite": "^5.4.8",
     "vitest": "^4.1.2"

tauri-app/pnpm-lock.yaml

@@ -45,15 +45,32 @@ importers:
       '@vitejs/plugin-react':
         specifier: ^4.3.1
         version: 4.7.0([email protected])
+      jsdom:
+        specifier: ^29.0.1
+        version: 29.0.1
       typescript:
         specifier: ^5.5.3
         version: 5.9.3
       vite:
         specifier: ^5.4.8
         version: 5.4.21
+      vitest:
+        specifier: ^4.1.2
+        version: 4.1.2([email protected])([email protected])
 
 packages:
 
+  '@asamuzakjp/[email protected]':
+    resolution: {integrity: sha512-iGWN8E45Ws0XWx3D44Q1t6vX2LqhCKcwfmwBYCDsFrYFS6m4q/Ks61L2veETaLv+ckDC6+dTETJoaAAb7VjLiw==}
+    engines: {node: ^20.19.0 || ^22.12.0 || >=24.0.0}
+
+  '@asamuzakjp/[email protected]':
+    resolution: {integrity: sha512-jXR6x4AcT3eIrS2fSNAwJpwirOkGcd+E7F7CP3zjdTqz9B/2huHOL8YJZBgekKwLML+u7qB/6P1LXQuMScsx0w==}
+    engines: {node: ^20.19.0 || ^22.12.0 || >=24.0.0}
+
+  '@asamuzakjp/[email protected]':
+    resolution: {integrity: sha512-n8GuYSrI9bF7FFZ/SjhwevlHc8xaVlb/7HmHelnc/PZXBD2ZR49NnN9sMMuDdEGPeeRQ5d0hqlSlEpgCX3Wl0Q==}
+
   '@babel/[email protected]':
     resolution: {integrity: sha512-9NhCeYjq9+3uxgdtp20LSiJXJvN0FeCtNGpJxuMFZ1Kv3cWUNb6DOhJwUvcVCzKGR66cw4njwM6hrJLqgOwbcw==}
     engines: {node: '>=6.9.0'}
@@ -137,6 +154,46 @@ packages:
     resolution: {integrity: sha512-LwdZHpScM4Qz8Xw2iKSzS+cfglZzJGvofQICy7W7v4caru4EaAmyUuO6BGrbyQ2mYV11W0U8j5mBhd14dd3B0A==}
     engines: {node: '>=6.9.0'}
 
+  '@bramus/[email protected]':
+    resolution: {integrity: sha512-ctxtJ/eA+t+6q2++vj5j7FYX3nRu311q1wfYH3xjlLOsczhlhxAg2FWNUXhpGvAw3BWo1xBcvOV6/YLc2r5FJw==}
+    hasBin: true
+
+  '@csstools/[email protected]':
+    resolution: {integrity: sha512-LMGQLS9EuADloEFkcTBR3BwV/CGHV7zyDxVRtVDTwdI2Ca4it0CCVTT9wCkxSgokjE5Ho41hEPgb8OEUwoXr6Q==}
+    engines: {node: '>=20.19.0'}
+
+  '@csstools/[email protected]':
+    resolution: {integrity: sha512-HJ26Z/vmsZQqs/o3a6bgKslXGFAungXGbinULZO3eMsOyNJHeBBZfup5FiZInOghgoM4Hwnmw+OgbJCNg1wwUQ==}
+    engines: {node: '>=20.19.0'}
+    peerDependencies:
+      '@csstools/css-parser-algorithms': ^4.0.0
+      '@csstools/css-tokenizer': ^4.0.0
+
+  '@csstools/[email protected]':
+    resolution: {integrity: sha512-0GEfbBLmTFf0dJlpsNU7zwxRIH0/BGEMuXLTCvFYxuL1tNhqzTbtnFICyJLTNK4a+RechKP75e7w42ClXSnJQw==}
+    engines: {node: '>=20.19.0'}
+    peerDependencies:
+      '@csstools/css-parser-algorithms': ^4.0.0
+      '@csstools/css-tokenizer': ^4.0.0
+
+  '@csstools/[email protected]':
+    resolution: {integrity: sha512-+B87qS7fIG3L5h3qwJ/IFbjoVoOe/bpOdh9hAjXbvx0o8ImEmUsGXN0inFOnk2ChCFgqkkGFQ+TpM5rbhkKe4w==}
+    engines: {node: '>=20.19.0'}
+    peerDependencies:
+      '@csstools/css-tokenizer': ^4.0.0
+
+  '@csstools/[email protected]':
+    resolution: {integrity: sha512-5GkLzz4prTIpoyeUiIu3iV6CSG3Plo7xRVOFPKI7FVEJ3mZ0A8SwK0XU3Gl7xAkiQ+mDyam+NNp875/C5y+jSA==}
+    peerDependencies:
+      css-tree: ^3.2.1
+    peerDependenciesMeta:
+      css-tree:
+        optional: true
+
+  '@csstools/[email protected]':
+    resolution: {integrity: sha512-QxULHAm7cNu72w97JUNCBFODFaXpbDg+dP8b/oWFAZ2MTRppA3U00Y2L1HqaS4J6yBqxwa/Y3nMBaxVKbB/NsA==}
+    engines: {node: '>=20.19.0'}
+
   '@esbuild/[email protected]':
     resolution: {integrity: sha512-1SDgH6ZSPTlggy1yI6+Dbkiz8xzpHJEVAlF/AM1tHPLsf5STom9rwtjE4hKAF20FfXXNTFqEYXyJNWh1GiZedQ==}
     engines: {node: '>=12'}
@@ -275,6 +332,15 @@ packages:
     cpu: [x64]
     os: [win32]
 
+  '@exodus/[email protected]':
+    resolution: {integrity: sha512-UY0nlA+feH81UGSHv92sLEPLCeZFjXOuHhrIo0HQydScuQc8s0A7kL/UdgwgDq8g8ilksmuoF35YVTNphV2aBQ==}
+    engines: {node: ^20.19.0 || ^22.12.0 || >=24.0.0}
+    peerDependencies:
+      '@noble/hashes': ^1.8.0 || ^2.0.0
+    peerDependenciesMeta:
+      '@noble/hashes':
+        optional: true
+
   '@jridgewell/[email protected]':
     resolution: {integrity: sha512-2kkt/7niJ6MgEPxF0bYdQ6etZaA+fQvDcLKckhy1yIQOzaoKjBBjSj63/aLVjYE3qhRt5dvM+uUyfCg6UKCBbA==}
 
@@ -432,6 +498,9 @@ packages:
     cpu: [x64]
     os: [win32]
 
+  '@standard-schema/[email protected]':
+    resolution: {integrity: sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w==}
+
   '@tauri-apps/[email protected]':
     resolution: {integrity: sha512-hKL/jWf293UDSUN09rR69hrToyIXBb8CjGaWC7gfinvnQrBVvnLr08FeFi38gxtugAVyVcTa5/FD/Xnkb1siBw==}
 
@@ -526,6 +595,12 @@ packages:
   '@types/[email protected]':
     resolution: {integrity: sha512-8PvcXf70gTDZBgt9ptxJ8elBeBjcLOAcOtoO/mPJjtji1+CdGbHgm77om1GrsPxsiE+uXIpNSK64UYaIwQXd4Q==}
 
+  '@types/[email protected]':
+    resolution: {integrity: sha512-Mw558oeA9fFbv65/y4mHtXDs9bPnFMZAL/jxdPFUpOHHIXX91mcgEHbS5Lahr+pwZFR8A7GQleRWeI6cGFC2UA==}
+
+  '@types/[email protected]':
+    resolution: {integrity: sha512-c9h9dVVMigMPc4bwTvC5dxqtqJZwQPePsWjPlpSOnojbor6pGqdk541lfA7AqFQr5pB1BRdq0juY9db81BwyFw==}
+
   '@types/[email protected]':
     resolution: {integrity: sha512-Fgg31wv9QbLDA0SpTOXO3MaxySc4DKGLi8sna4/Utjo4r3ZRPdCt4UQee8BWr+Q5z21yifghREPJGYaEOEIACg==}
     deprecated: This is a stub types definition. dompurify provides its own type definitions, so you do not need this installed.
@@ -553,11 +628,47 @@ packages:
     peerDependencies:
       vite: ^4.2.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
 
+  '@vitest/[email protected]':
+    resolution: {integrity: sha512-gbu+7B0YgUJ2nkdsRJrFFW6X7NTP44WlhiclHniUhxADQJH5Szt9mZ9hWnJPJ8YwOK5zUOSSlSvyzRf0u1DSBQ==}
+
+  '@vitest/[email protected]':
+    resolution: {integrity: sha512-Ize4iQtEALHDttPRCmN+FKqOl2vxTiNUhzobQFFt/BM1lRUTG7zRCLOykG/6Vo4E4hnUdfVLo5/eqKPukcWW7Q==}
+    peerDependencies:
+      msw: ^2.4.9
+      vite: ^6.0.0 || ^7.0.0 || ^8.0.0
+    peerDependenciesMeta:
+      msw:
+        optional: true
+      vite:
+        optional: true
+
+  '@vitest/[email protected]':
+    resolution: {integrity: sha512-dwQga8aejqeuB+TvXCMzSQemvV9hNEtDDpgUKDzOmNQayl2OG241PSWeJwKRH3CiC+sESrmoFd49rfnq7T4RnA==}
+
+  '@vitest/[email protected]':
+    resolution: {integrity: sha512-Gr+FQan34CdiYAwpGJmQG8PgkyFVmARK8/xSijia3eTFgVfpcpztWLuP6FttGNfPLJhaZVP/euvujeNYar36OQ==}
+
+  '@vitest/[email protected]':
+    resolution: {integrity: sha512-g7yfUmxYS4mNxk31qbOYsSt2F4m1E02LFqO53Xpzg3zKMhLAPZAjjfyl9e6z7HrW6LvUdTwAQR3HHfLjpko16A==}
+
+  '@vitest/[email protected]':
+    resolution: {integrity: sha512-DU4fBnbVCJGNBwVA6xSToNXrkZNSiw59H8tcuUspVMsBDBST4nfvsPsEHDHGtWRRnqBERBQu7TrTKskmjqTXKA==}
+
+  '@vitest/[email protected]':
+    resolution: {integrity: sha512-xw2/TiX82lQHA06cgbqRKFb5lCAy3axQ4H4SoUFhUsg+wztiet+co86IAMDtF6Vm1hc7J6j09oh/rgDn+JdKIQ==}
+
+  [email protected]:
+    resolution: {integrity: sha512-Izi8RQcffqCeNVgFigKli1ssklIbpHnCYc6AknXGYoB6grJqyeby7jv12JUQgmTAnIDnbck1uxksT4dzN3PWBA==}
+    engines: {node: '>=12'}
+
   [email protected]:
     resolution: {integrity: sha512-DAKrHphkJyiGuau/cFieRYhcTFeK/lBuD++C7cZ6KZHbMhBrisoi+EvhQ5RZrIfV5qwsW8kgQ07JIC+MDJRAhg==}
     engines: {node: '>=6.0.0'}
     hasBin: true
 
+  [email protected]:
+    resolution: {integrity: sha512-RKshQI1R3YQ+n9YJz2QQ147P66ELpa1FQEg20Dk8oW9t2KgLbpDLLp9aGZ7y8WHSshDknG0bknqGw5/tyCs5tw==}
+
   [email protected]:
     resolution: {integrity: sha512-ZC5Bd0LgJXgwGqUknZY/vkUQ04r8NXnJZ3yYi4vDmSiZmC/pdSN0NbNRPxZpbtO4uAfDUAFffO8IZoM3Gj8IkA==}
     engines: {node: ^6 || ^7 || ^8 || ^9 || ^10 || ^11 || ^12 || >=13.7}
@@ -566,12 +677,24 @@ packages:
   [email protected]:
     resolution: {integrity: sha512-RdwNCyMsNBftLjW6w01z8bKEvT6e/5tpPVEgtn22TiLGlstHOVecsX2KHFkD5e/vRnIE4EGzpuIODb3mtswtkw==}
 
+  [email protected]:
+    resolution: {integrity: sha512-NUPRluOfOiTKBKvWPtSD4PhFvWCqOi0BGStNWs57X9js7XGTprSmFoz5F0tWhR4WPjNeR9jXqdC7/UpSJTnlRg==}
+    engines: {node: '>=18'}
+
   [email protected]:
     resolution: {integrity: sha512-Kvp459HrV2FEJ1CAsi1Ku+MY3kasH19TFykTz2xWmMeq6bk2NU3XXvfJ+Q61m0xktWwt+1HSYf3JZsTms3aRJg==}
 
+  [email protected]:
+    resolution: {integrity: sha512-X7sjQzceUhu1u7Y/ylrRZFU2FS6LRiFVp6rKLPg23y3x3c3DOKAwuXGDp+PAGjh6CSnCjYeAul8pcT8bAl+lSA==}
+    engines: {node: ^10 || ^12.20.0 || ^14.13.0 || >=15.0.0}
+
   [email protected]:
     resolution: {integrity: sha512-z1HGKcYy2xA8AGQfwrn0PAy+PB7X/GSj3UVJW9qKyn43xWa+gl5nXmU4qqLMRzWVLFC8KusUX8T/0kCiOYpAIQ==}
 
+  [email protected]:
+    resolution: {integrity: sha512-23XHcCF+coGYevirZceTVD7NdJOqVn+49IHyxgszm+JIiHLoB2TkmPtsYkNWT1pvRSGkc35L6NHs0yHkN2SumA==}
+    engines: {node: ^20.19.0 || ^22.12.0 || >=24.0.0}
+
   [email protected]:
     resolution: {integrity: sha512-RGwwWnwQvkVfavKVt22FGLw+xYSdzARwm0ru6DhTVA3umU5hZc28V3kO4stgYryrTlLpuvgI9GiijltAjNbcqA==}
     engines: {node: '>=6.0'}
@@ -581,12 +704,22 @@ packages:
       supports-color:
         optional: true
 
+  [email protected]:
+    resolution: {integrity: sha512-YpgQiITW3JXGntzdUmyUR1V812Hn8T1YVXhCu+wO3OpS4eU9l4YdD3qjyiKdV6mvV29zapkMeD390UVEf2lkUg==}
+
   [email protected]:
     resolution: {integrity: sha512-Oj6pzI2+RqBfFG+qOaOLbFXLQ90ARpcGG6UePL82bJLtdsa6CYJD7nmiU8MW9nQNOtCHV3lZ/Bzq1X0QYbBZCA==}
 
   [email protected]:
     resolution: {integrity: sha512-uRBlUfKKdsXMkiiOurgaybNC10tjrD+skXLEg7NHbm6h0uAoqj3xMb9uue5BfcSCXJ4mcyJMOucI6q55D7p6KQ==}
 
+  [email protected]:
+    resolution: {integrity: sha512-aN97NXWF6AWBTahfVOIrB/NShkzi5H7F9r1s9mD3cDj4Ko5f2qhhVoYMibXF7GlLveb/D2ioWay8lxI97Ven3g==}
+    engines: {node: '>=0.12'}
+
+  [email protected]:
+    resolution: {integrity: sha512-5POEcUuZybH7IdmGsD8wlf0AI55wMecM9rVBTI/qEAy2c1kTOm3DjFYjrBdI2K3BaJjJYfYFeRtM0t9ssnRuxw==}
+
   [email protected]:
     resolution: {integrity: sha512-mg3OPMV4hXywwpoDxu3Qda5xCKQi+vCTZq8S9J/EpkhB2HzKXq4SNFZE3+NK93JYxc8VMSep+lOUSC/RVKaBqw==}
     engines: {node: '>=12'}
@@ -596,6 +729,22 @@ packages:
     resolution: {integrity: sha512-WUj2qlxaQtO4g6Pq5c29GTcWGDyd8itL8zTlipgECz3JesAiiOKotd8JU6otB3PACgG6xkJUyVhboMS+bje/jA==}
     engines: {node: '>=6'}
 
+  [email protected]:
+    resolution: {integrity: sha512-7RUKfXgSMMkzt6ZuXmqapOurLGPPfgj6l9uRZ7lRGolvk0y2yocc35LdcxKC5PQZdn2DMqioAQ2NoWcrTKmm6g==}
+
+  [email protected]:
+    resolution: {integrity: sha512-knvyeauYhqjOYvQ66MznSMs83wmHrCycNEN6Ao+2AeYEfxUIkuiVxdEa1qlGEPK+We3n0THiDciYSsCcgW/DoA==}
+    engines: {node: '>=12.0.0'}
+
+  [email protected]:
+    resolution: {integrity: sha512-tIbYtZbucOs0BRGqPJkshJUYdL+SDH7dVM8gjy+ERp3WAUjLEFJE+02kanyHtwjWOnwrKYBiwAmM0p4kLJAnXg==}
+    engines: {node: '>=12.0.0'}
+    peerDependencies:
+      picomatch: ^3 || ^4
+    peerDependenciesMeta:
+      picomatch:
+        optional: true
+
   [email protected]:
     resolution: {integrity: sha512-5xoDfX+fL7faATnagmWPpbFtwh/R77WmMMqqHGS65C3vvB0YHrgF+B1YmZ3441tMj5n63k0212XNoJwzlhffQw==}
     engines: {node: ^8.16.0 || ^10.6.0 || >=11.0.0}
@@ -609,9 +758,25 @@ packages:
     resolution: {integrity: sha512-Xwwo44whKBVCYoliBQwaPvtd/2tYFkRQtXDWj1nackaV2JPXx3L0+Jvd8/qCJ2p+ML0/XVkJ2q+Mr+UVdpJK5w==}
     engines: {node: '>=12.0.0'}
 
+  [email protected]:
+    resolution: {integrity: sha512-CV9TW3Y3f8/wT0BRFc1/KAVQ3TUHiXmaAb6VW9vtiMFf7SLoMd1PdAc4W3KFOFETBJUb90KatHqlsZMWV+R9Gg==}
+    engines: {node: ^20.19.0 || ^22.12.0 || >=24.0.0}
+
+  [email protected]:
+    resolution: {integrity: sha512-bCYeRA2rVibKZd+s2625gGnGF/t7DSqDs4dP7CrLA1m7jKWz6pps0LpYLJN8Q64HtmPKJ1hrN3nzPNKFEKOUiQ==}
+
   [email protected]:
     resolution: {integrity: sha512-RdJUflcE3cUzKiMqQgsCu06FPu9UdIJO0beYbPhHN4k6apgJtifcoCtT9bcxOpYBtpD2kCM6Sbzg4CausW/PKQ==}
 
+  [email protected]:
+    resolution: {integrity: sha512-z6JOK5gRO7aMybVq/y/MlIpKh8JIi68FBKMUtKkK2KH/wMSRlCxQ682d08LB9fYXplyY/UXG8P4XXTScmdjApg==}
+    engines: {node: ^20.19.0 || ^22.13.0 || >=24.0.0}
+    peerDependencies:
+      canvas: ^3.0.0
+    peerDependenciesMeta:
+      canvas:
+        optional: true
+
   [email protected]:
     resolution: {integrity: sha512-/sM3dO2FOzXjKQhJuo0Q173wf2KOo8t4I8vHy6lF9poUp7bKT0/NHE8fPX23PwfhnykfqnC2xRxOnVw5XuGIaA==}
     engines: {node: '>=6'}
@@ -626,14 +791,24 @@ packages:
     resolution: {integrity: sha512-lyuxPGr/Wfhrlem2CL/UcnUc1zcqKAImBDzukY7Y5F/yQiNdko6+fRLevlw1HgMySw7f611UIY408EtxRSoK3Q==}
     hasBin: true
 
+  [email protected]:
+    resolution: {integrity: sha512-aY/R+aEsRelme17KGQa/1ZSIpLpNYYrhcrepKTZgE+W3WM16YMCaPwOHLHsmopZHELU0Ojin1lPVxKR0MihncA==}
+    engines: {node: 20 || >=22}
+
   [email protected]:
     resolution: {integrity: sha512-KpNARQA3Iwv+jTA0utUVVbrh+Jlrr1Fv0e56GGzAFOXN7dk/FviaDW8LHmK52DlcH4WP2n6gI8vN1aesBFgo9w==}
 
+  [email protected]:
+    resolution: {integrity: sha512-vd2F4YUyEXKGcLHoq+TEyCjxueSeHnFxyyjNp80yg0XV4vUhnDer/lvvlqM/arB5bXQN5K2/3oinyCRyx8T2CQ==}
+
   [email protected]:
     resolution: {integrity: sha512-qXUm7e/YKFoqFPYPa3Ukg9xlI5cyAtGmyEIzMfW//m6kXwCy2Ps9DYf5ioijFKQ8qyuscrHoY04iJGctu2Kg0Q==}
     engines: {node: '>= 18'}
     hasBin: true
 
+  [email protected]:
+    resolution: {integrity: sha512-9Yubnt3e8A0OKwxYSXyhLymGW4sCufcLG6VdiDdUGVkPhpqLxlvP5vl1983gQjJl3tqbrM731mjaZaP68AgosQ==}
+
   [email protected]:
     resolution: {integrity: sha512-6FlzubTLZG3J2a/NVCAleEhjzq5oxgHyaCU9yYXvcLsvoVaHJq/s5xXI6/XXP6tz7R9xAOtHnSO/tXtF3WRTlA==}
 
@@ -645,13 +820,30 @@ packages:
   [email protected]:
     resolution: {integrity: sha512-TdC8FSgHz8Mwtw9g5L4gR/Sh9XhSP/0DEkQxfEFXOpiul5IiHgHan2VhYYb6agDSfp4KuvltmGApc8HMgUrIkA==}
 
+  [email protected]:
+    resolution: {integrity: sha512-uTqF9MuPraAQ+IsnPf366RG4cP9RtUi7MLO1N3KEc+wb0a6yKpeL0lmk2IB1jY5KHPAlTc6T/JRdC/YqxHNwkQ==}
+
+  [email protected]:
+    resolution: {integrity: sha512-9m4m5GSgXjL4AjumKzq1Fgfp3Z8rsvjRNbnkVwfu2ImRqE5D0LnY2QfDen18FSY9C573YU5XxSapdHZTZ2WolA==}
+
+  [email protected]:
+    resolution: {integrity: sha512-WUjGcAqP1gQacoQe+OBJsFA7Ld4DyXuUIjZ5cc75cLHvJ7dtNsTugphxIADwspS+AraAUePCKrSVtPLFj/F88w==}
+
   [email protected]:
     resolution: {integrity: sha512-xceH2snhtb5M9liqDsmEw56le376mTZkEX/jEb/RxNFyegNul7eNslCXP9FDj/Lcu0X8KEyMceP2ntpaHrDEVA==}
 
+  [email protected]:
+    resolution: {integrity: sha512-QP88BAKvMam/3NxH6vj2o21R6MjxZUAd6nlwAS/pnGvN9IVLocLHxGYIzFhg6fUQ+5th6P4dv4eW9jX3DSIj7A==}
+    engines: {node: '>=12'}
+
   [email protected]:
     resolution: {integrity: sha512-OW/rX8O/jXnm82Ey1k44pObPtdblfiuWnrd8X7GJ7emImCOstunGbXUpp7HdBrFQX6rJzn3sPT397Wp5aCwCHg==}
     engines: {node: ^10 || ^12 || >=14}
 
+  [email protected]:
+    resolution: {integrity: sha512-vYt7UD1U9Wg6138shLtLOvdAu+8DsC/ilFtEVHcH+wydcSpNE20AfSOduf6MkRFahL5FY7X1oU7nKVZFtfq8Fg==}
+    engines: {node: '>=6'}
+
   [email protected]:
     resolution: {integrity: sha512-5m4nQKp+rZRb09LNH59GM4BxTh9251/ylbKIbpe7TpGxfJ+9kv6BLkLBXIjjspbgbnIBNqlI23tRnTWT0snUIw==}
     peerDependencies:
@@ -665,11 +857,19 @@ packages:
     resolution: {integrity: sha512-wS+hAgJShR0KhEvPJArfuPVN1+Hz1t0Y6n5jLrGQbkb4urgPE/0Rve+1kMB1v/oWgHgm4WIcV+i7F2pTVj+2iQ==}
     engines: {node: '>=0.10.0'}
 
+  [email protected]:
+    resolution: {integrity: sha512-Xf0nWe6RseziFMu+Ap9biiUbmplq6S9/p+7w7YXP/JBHhrUDDUhwa+vANyubuqfZWTveU//DYVGsDG7RKL/vEw==}
+    engines: {node: '>=0.10.0'}
+
   [email protected]:
     resolution: {integrity: sha512-yqjxruMGBQJ2gG4HtjZtAfXArHomazDHoFwFFmZZl0r7Pdo7qCIXKqKHZc8yeoMgzJJ+pO6pEEHa+V7uzWlrAQ==}
     engines: {node: '>=18.0.0', npm: '>=8.0.0'}
     hasBin: true
 
+  [email protected]:
+    resolution: {integrity: sha512-xAg7SOnEhrm5zI3puOOKyy1OMcMlIJZYNJY7xLBwSze0UjhPLnWfj2GF2EpT0jmzaJKIWKHLsaSSajf35bcYnA==}
+    engines: {node: '>=v12.22.7'}
+
   [email protected]:
     resolution: {integrity: sha512-UOShsPwz7NrMUqhR6t0hWjFduvOzbtv7toDH1/hIrfRNIDBnnBWd0CwJTGvTpngVlmwGCdP9/Zl/tVrDqcuYzQ==}
 
@@ -677,15 +877,61 @@ packages:
     resolution: {integrity: sha512-BR7VvDCVHO+q2xBEWskxS6DJE1qRnb7DxzUrogb71CWoSficBxYsiAGd+Kl0mmq/MprG9yArRkyrQxTO6XjMzA==}
     hasBin: true
 
+  [email protected]:
+    resolution: {integrity: sha512-ybx0WO1/8bSBLEWXZvEd7gMW3Sn3JFlW3TvX1nREbDLRNQNaeNN8WK0meBwPdAaOI7TtRRRJn/Es1zhrrCHu7g==}
+
   [email protected]:
     resolution: {integrity: sha512-UXWMKhLOwVKb728IUtQPXxfYU+usdybtUrK/8uGE8CQMvrhOpwvzDBwj0QhSL7MQc7vIsISBG8VQ8+IDQxpfQA==}
     engines: {node: '>=0.10.0'}
 
+  [email protected]:
+    resolution: {integrity: sha512-1XMJE5fQo1jGH6Y/7ebnwPOBEkIEnT4QF32d5R1+VXdXveM0IBMJt8zfaxX1P3QhVwrYe+576+jkANtSS2mBbw==}
+
+  [email protected]:
+    resolution: {integrity: sha512-zUMPtQ/HBY3/50VbpkupYHbRroTRZJPRLvreamgErJVys0ceuzMkD44J/QjqhHjOzK42GQ3QZIeFG1OYfOtKqQ==}
+
+  [email protected]:
+    resolution: {integrity: sha512-9QNk5KwDF+Bvz+PyObkmSYjI5ksVUYtjW7AU22r2NKcfLJcXp96hkDWU3+XndOsUb+AQ9QhfzfCT2O+CNWT5Tw==}
+
+  [email protected]:
+    resolution: {integrity: sha512-0+DUvqWMValLmha6lr4kD8iAMK1HzV0/aKnCtWb9v9641TnP/MFb7Pc2bxoxQjTXAErryXVgUOfv2YqNllqGeg==}
+
+  [email protected]:
+    resolution: {integrity: sha512-u9r3uZC0bdpGOXtlxUIdwf9pkmvhqJdrVCH9fapQtgy/OeTTMZ1nqH7agtvEfmGui6e1XxjcdrlxvxJvc3sMqw==}
+    engines: {node: '>=18'}
+
+  [email protected]:
+    resolution: {integrity: sha512-j2Zq4NyQYG5XMST4cbs02Ak8iJUdxRM0XI5QyxXuZOzKOINmWurp3smXu3y5wDcJrptwpSjgXHzIQxR0omXljQ==}
+    engines: {node: '>=12.0.0'}
+
+  [email protected]:
+    resolution: {integrity: sha512-Bf+ILmBgretUrdJxzXM0SgXLZ3XfiaUuOj/IKQHuTXip+05Xn+uyEYdVg0kYDipTBcLrCVyUzAPz7QmArb0mmw==}
+    engines: {node: '>=14.0.0'}
+
+  [email protected]:
+    resolution: {integrity: sha512-YQ7uPjgWUibIK6DW5lrKujGwUKhLevU4hcGbP5O6TcIUb+oTjJYJVWPS4nZsIHrEEEG6myk/oqAJUEQmpZrHsg==}
+
+  [email protected]:
+    resolution: {integrity: sha512-I4FZcVFcqCRuT0ph6dCDpPuO4Xgzvh+spkcTr1gK7peIvxWauoloVO0vuy1FQnijT63ss6AsHB6+OIM4aXHbPg==}
+    hasBin: true
+
+  [email protected]:
+    resolution: {integrity: sha512-LktZQb3IeoUWB9lqR5EWTHgW/VTITCXg4D21M+lvybRVdylLrRMnqaIONLVb5mav8vM19m44HIcGq4qASeu2Qw==}
+    engines: {node: '>=16'}
+
+  [email protected]:
+    resolution: {integrity: sha512-bLVMLPtstlZ4iMQHpFHTR7GAGj2jxi8Dg0s2h2MafAE4uSWF98FC/3MomU51iQAMf8/qDUbKWf5GxuvvVcXEhw==}
+    engines: {node: '>=20'}
+
   [email protected]:
     resolution: {integrity: sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==}
     engines: {node: '>=14.17'}
     hasBin: true
 
+  [email protected]:
+    resolution: {integrity: sha512-Xi4agocCbRzt0yYMZGMA6ApD7gvtUFaxm4ZmeacWI4cZxaF6C+8I8QfofC20NAePiB/IcvZmzkJ7XPa471AEtA==}
+    engines: {node: '>=20.18.1'}
+
   [email protected]:
     resolution: {integrity: sha512-Js0m9cx+qOgDxo0eMiFGEueWztz+d4+M3rGlmKPT+T4IS/jP4ylw3Nwpu6cpTTP8R1MAC1kF4VbdLt3ARf209w==}
     hasBin: true
@@ -723,11 +969,92 @@ packages:
       terser:
         optional: true
 
+  [email protected]:
+    resolution: {integrity: sha512-xjR1dMTVHlFLh98JE3i/f/WePqJsah4A0FK9cc8Ehp9Udk0AZk6ccpIZhh1qJ/yxVWRZ+Q54ocnD8TXmkhspGg==}
+    engines: {node: ^20.0.0 || ^22.0.0 || >=24.0.0}
+    hasBin: true
+    peerDependencies:
+      '@edge-runtime/vm': '*'
+      '@opentelemetry/api': ^1.9.0
+      '@types/node': ^20.0.0 || ^22.0.0 || >=24.0.0
+      '@vitest/browser-playwright': 4.1.2
+      '@vitest/browser-preview': 4.1.2
+      '@vitest/browser-webdriverio': 4.1.2
+      '@vitest/ui': 4.1.2
+      happy-dom: '*'
+      jsdom: '*'
+      vite: ^6.0.0 || ^7.0.0 || ^8.0.0
+    peerDependenciesMeta:
+      '@edge-runtime/vm':
+        optional: true
+      '@opentelemetry/api':
+        optional: true
+      '@types/node':
+        optional: true
+      '@vitest/browser-playwright':
+        optional: true
+      '@vitest/browser-preview':
+        optional: true
+      '@vitest/browser-webdriverio':
+        optional: true
+      '@vitest/ui':
+        optional: true
+      happy-dom:
+        optional: true
+      jsdom:
+        optional: true
+
+  [email protected]:
+    resolution: {integrity: sha512-o8qghlI8NZHU1lLPrpi2+Uq7abh4GGPpYANlalzWxyWteJOCsr/P+oPBA49TOLu5FTZO4d3F9MnWJfiMo4BkmA==}
+    engines: {node: '>=18'}
+
+  [email protected]:
+    resolution: {integrity: sha512-BMhLD/Sw+GbJC21C/UgyaZX41nPt8bUTg+jWyDeg7e7YN4xOM05YPSIXceACnXVtqyEw/LMClUQMtMZ+PGGpqQ==}
+    engines: {node: '>=20'}
+
+  [email protected]:
+    resolution: {integrity: sha512-sXcNcHOC51uPGF0P/D4NVtrkjSU2fNsm9iog4ZvZJsL3rjoDAzXZhkm2MWt1y+PUdggKAYVoMAIYcs78wJ51Cw==}
+    engines: {node: '>=20'}
+
+  [email protected]:
+    resolution: {integrity: sha512-1to4zXBxmXHV3IiSSEInrreIlu02vUOvrhxJJH5vcxYTBDAx51cqZiKdyTxlecdKNSjj8EcxGBxNf6Vg+945gw==}
+    engines: {node: ^20.19.0 || ^22.12.0 || >=24.0.0}
+
+  [email protected]:
+    resolution: {integrity: sha512-hUrmaWBdVDcxvYqnyh09zunKzROWjbZTiNy8dBEjkS7ehEDQibXJ7XvlmtbwuTclUiIyN+CyXQD4Vmko8fNm8w==}
+    engines: {node: '>=8'}
+    hasBin: true
+
+  [email protected]:
+    resolution: {integrity: sha512-EvGK8EJ3DhaHfbRlETOWAS5pO9MZITeauHKJyb8wyajUfQUenkIg2MvLDTZ4T/TgIcm3HU0TFBgWWboAZ30UHg==}
+    engines: {node: '>=18'}
+
+  [email protected]:
+    resolution: {integrity: sha512-JZnDKK8B0RCDw84FNdDAIpZK+JuJw+s7Lz8nksI7SIuU3UXJJslUthsi+uWBUYOwPFwW7W7PRLRfUKpxjtjFCw==}
+
   [email protected]:
     resolution: {integrity: sha512-a4UGQaWPH59mOXUYnAG2ewncQS4i4F43Tv3JoAM+s2VDAmS9NsK8GpDMLrCHPksFT7h3K6TOoUNn2pb7RoXx4g==}
 
 snapshots:
 
+  '@asamuzakjp/[email protected]':
+    dependencies:
+      '@csstools/css-calc': 3.1.1(@csstools/[email protected](@csstools/[email protected]))(@csstools/[email protected])
+      '@csstools/css-color-parser': 4.0.2(@csstools/[email protected](@csstools/[email protected]))(@csstools/[email protected])
+      '@csstools/css-parser-algorithms': 4.0.0(@csstools/[email protected])
+      '@csstools/css-tokenizer': 4.0.0
+      lru-cache: 11.2.7
+
+  '@asamuzakjp/[email protected]':
+    dependencies:
+      '@asamuzakjp/nwsapi': 2.3.9
+      bidi-js: 1.0.3
+      css-tree: 3.2.1
+      is-potential-custom-element-name: 1.0.1
+      lru-cache: 11.2.7
+
+  '@asamuzakjp/[email protected]': {}
+
   '@babel/[email protected]':
     dependencies:
       '@babel/helper-validator-identifier': 7.28.5
@@ -840,6 +1167,34 @@ snapshots:
       '@babel/helper-string-parser': 7.27.1
       '@babel/helper-validator-identifier': 7.28.5
 
+  '@bramus/[email protected]':
+    dependencies:
+      css-tree: 3.2.1
+
+  '@csstools/[email protected]': {}
+
+  '@csstools/[email protected](@csstools/[email protected](@csstools/[email protected]))(@csstools/[email protected])':
+    dependencies:
+      '@csstools/css-parser-algorithms': 4.0.0(@csstools/[email protected])
+      '@csstools/css-tokenizer': 4.0.0
+
+  '@csstools/[email protected](@csstools/[email protected](@csstools/[email protected]))(@csstools/[email protected])':
+    dependencies:
+      '@csstools/color-helpers': 6.0.2
+      '@csstools/css-calc': 3.1.1(@csstools/[email protected](@csstools/[email protected]))(@csstools/[email protected])
+      '@csstools/css-parser-algorithms': 4.0.0(@csstools/[email protected])
+      '@csstools/css-tokenizer': 4.0.0
+
+  '@csstools/[email protected](@csstools/[email protected])':
+    dependencies:
+      '@csstools/css-tokenizer': 4.0.0
+
+  '@csstools/[email protected]([email protected])':
+    optionalDependencies:
+      css-tree: 3.2.1
+
+  '@csstools/[email protected]': {}
+
   '@esbuild/[email protected]':
     optional: true
 
@@ -909,6 +1264,8 @@ snapshots:
   '@esbuild/[email protected]':
     optional: true
 
+  '@exodus/[email protected]': {}
+
   '@jridgewell/[email protected]':
     dependencies:
       '@jridgewell/sourcemap-codec': 1.5.5
@@ -1005,6 +1362,8 @@ snapshots:
   '@rollup/[email protected]':
     optional: true
 
+  '@standard-schema/[email protected]': {}
+
   '@tauri-apps/[email protected]': {}
 
   '@tauri-apps/[email protected]':
@@ -1079,6 +1438,13 @@ snapshots:
     dependencies:
       '@babel/types': 7.29.0
 
+  '@types/[email protected]':
+    dependencies:
+      '@types/deep-eql': 4.0.2
+      assertion-error: 2.0.1
+
+  '@types/[email protected]': {}
+
   '@types/[email protected]':
     dependencies:
       dompurify: 3.3.3
@@ -1111,8 +1477,55 @@ snapshots:
     transitivePeerDependencies:
       - supports-color
 
+  '@vitest/[email protected]':
+    dependencies:
+      '@standard-schema/spec': 1.1.0
+      '@types/chai': 5.2.3
+      '@vitest/spy': 4.1.2
+      '@vitest/utils': 4.1.2
+      chai: 6.2.2
+      tinyrainbow: 3.1.0
+
+  '@vitest/[email protected]([email protected])':
+    dependencies:
+      '@vitest/spy': 4.1.2
+      estree-walker: 3.0.3
+      magic-string: 0.30.21
+    optionalDependencies:
+      vite: 5.4.21
+
+  '@vitest/[email protected]':
+    dependencies:
+      tinyrainbow: 3.1.0
+
+  '@vitest/[email protected]':
+    dependencies:
+      '@vitest/utils': 4.1.2
+      pathe: 2.0.3
+
+  '@vitest/[email protected]':
+    dependencies:
+      '@vitest/pretty-format': 4.1.2
+      '@vitest/utils': 4.1.2
+      magic-string: 0.30.21
+      pathe: 2.0.3
+
+  '@vitest/[email protected]': {}
+
+  '@vitest/[email protected]':
+    dependencies:
+      '@vitest/pretty-format': 4.1.2
+      convert-source-map: 2.0.0
+      tinyrainbow: 3.1.0
+
+  [email protected]: {}
+
   [email protected]: {}
 
+  [email protected]:
+    dependencies:
+      require-from-string: 2.0.2
+
   [email protected]:
     dependencies:
       baseline-browser-mapping: 2.10.11
@@ -1123,20 +1536,40 @@ snapshots:
 
   [email protected]: {}
 
+  [email protected]: {}
+
   [email protected]: {}
 
+  [email protected]:
+    dependencies:
+      mdn-data: 2.27.1
+      source-map-js: 1.2.1
+
   [email protected]: {}
 
+  [email protected]:
+    dependencies:
+      whatwg-mimetype: 5.0.0
+      whatwg-url: 16.0.1
+    transitivePeerDependencies:
+      - '@noble/hashes'
+
   [email protected]:
     dependencies:
       ms: 2.1.3
 
+  [email protected]: {}
+
   [email protected]:
     optionalDependencies:
       '@types/trusted-types': 2.0.7
 
   [email protected]: {}
 
+  [email protected]: {}
+
+  [email protected]: {}
+
   [email protected]:
     optionalDependencies:
       '@esbuild/aix-ppc64': 0.21.5
@@ -1165,6 +1598,16 @@ snapshots:
 
   [email protected]: {}
 
+  [email protected]:
+    dependencies:
+      '@types/estree': 1.0.8
+
+  [email protected]: {}
+
+  [email protected]([email protected]):
+    optionalDependencies:
+      picomatch: 4.0.4
+
   [email protected]:
     optional: true
 
@@ -1172,8 +1615,42 @@ snapshots:
 
   [email protected]: {}
 
+  [email protected]:
+    dependencies:
+      '@exodus/bytes': 1.15.0
+    transitivePeerDependencies:
+      - '@noble/hashes'
+
+  [email protected]: {}
+
   [email protected]: {}
 
+  [email protected]:
+    dependencies:
+      '@asamuzakjp/css-color': 5.1.1
+      '@asamuzakjp/dom-selector': 7.0.4
+      '@bramus/specificity': 2.4.2
+      '@csstools/css-syntax-patches-for-csstree': 1.1.2([email protected])
+      '@exodus/bytes': 1.15.0
+      css-tree: 3.2.1
+      data-urls: 7.0.0
+      decimal.js: 10.6.0
+      html-encoding-sniffer: 6.0.0
+      is-potential-custom-element-name: 1.0.1
+      lru-cache: 11.2.7
+      parse5: 8.0.0
+      saxes: 6.0.0
+      symbol-tree: 3.2.4
+      tough-cookie: 6.0.1
+      undici: 7.24.6
+      w3c-xmlserializer: 5.0.0
+      webidl-conversions: 8.0.1
+      whatwg-mimetype: 5.0.0
+      whatwg-url: 16.0.1
+      xml-name-validator: 5.0.0
+    transitivePeerDependencies:
+      - '@noble/hashes'
+
   [email protected]: {}
 
   [email protected]: {}
@@ -1182,26 +1659,46 @@ snapshots:
     dependencies:
       js-tokens: 4.0.0
 
+  [email protected]: {}
+
   [email protected]:
     dependencies:
       yallist: 3.1.1
 
+  [email protected]:
+    dependencies:
+      '@jridgewell/sourcemap-codec': 1.5.5
+
   [email protected]: {}
 
+  [email protected]: {}
+
   [email protected]: {}
 
   [email protected]: {}
 
   [email protected]: {}
 
+  [email protected]: {}
+
+  [email protected]:
+    dependencies:
+      entities: 6.0.1
+
+  [email protected]: {}
+
   [email protected]: {}
 
+  [email protected]: {}
+
   [email protected]:
     dependencies:
       nanoid: 3.3.11
       picocolors: 1.1.1
       source-map-js: 1.2.1
 
+  [email protected]: {}
+
   [email protected]([email protected]):
     dependencies:
       loose-envify: 1.4.0
@@ -1214,6 +1711,8 @@ snapshots:
     dependencies:
       loose-envify: 1.4.0
 
+  [email protected]: {}
+
   [email protected]:
     dependencies:
       '@types/estree': 1.0.8
@@ -1245,16 +1744,55 @@ snapshots:
       '@rollup/rollup-win32-x64-msvc': 4.60.0
       fsevents: 2.3.3
 
+  [email protected]:
+    dependencies:
+      xmlchars: 2.2.0
+
   [email protected]:
     dependencies:
       loose-envify: 1.4.0
 
   [email protected]: {}
 
+  [email protected]: {}
+
   [email protected]: {}
 
+  [email protected]: {}
+
+  [email protected]: {}
+
+  [email protected]: {}
+
+  [email protected]: {}
+
+  [email protected]: {}
+
+  [email protected]:
+    dependencies:
+      fdir: 6.5.0([email protected])
+      picomatch: 4.0.4
+
+  [email protected]: {}
+
+  [email protected]: {}
+
+  [email protected]:
+    dependencies:
+      tldts-core: 7.0.27
+
+  [email protected]:
+    dependencies:
+      tldts: 7.0.27
+
+  [email protected]:
+    dependencies:
+      punycode: 2.3.1
+
   [email protected]: {}
 
+  [email protected]: {}
+
   [email protected]([email protected]):
     dependencies:
       browserslist: 4.28.1
@@ -1269,4 +1807,56 @@ snapshots:
     optionalDependencies:
       fsevents: 2.3.3
 
+  [email protected]([email protected])([email protected]):
+    dependencies:
+      '@vitest/expect': 4.1.2
+      '@vitest/mocker': 4.1.2([email protected])
+      '@vitest/pretty-format': 4.1.2
+      '@vitest/runner': 4.1.2
+      '@vitest/snapshot': 4.1.2
+      '@vitest/spy': 4.1.2
+      '@vitest/utils': 4.1.2
+      es-module-lexer: 2.0.0
+      expect-type: 1.3.0
+      magic-string: 0.30.21
+      obug: 2.1.1
+      pathe: 2.0.3
+      picomatch: 4.0.4
+      std-env: 4.0.0
+      tinybench: 2.9.0
+      tinyexec: 1.0.4
+      tinyglobby: 0.2.15
+      tinyrainbow: 3.1.0
+      vite: 5.4.21
+      why-is-node-running: 2.3.0
+    optionalDependencies:
+      jsdom: 29.0.1
+    transitivePeerDependencies:
+      - msw
+
+  [email protected]:
+    dependencies:
+      xml-name-validator: 5.0.0
+
+  [email protected]: {}
+
+  [email protected]: {}
+
+  [email protected]:
+    dependencies:
+      '@exodus/bytes': 1.15.0
+      tr46: 6.0.0
+      webidl-conversions: 8.0.1
+    transitivePeerDependencies:
+      - '@noble/hashes'
+
+  [email protected]:
+    dependencies:
+      siginfo: 2.0.0
+      stackback: 0.0.2
+
+  [email protected]: {}
+
+  [email protected]: {}
+
   [email protected]: {}

tauri-app/postcss.config.js

@@ -0,0 +1,6 @@
+export default {
+  plugins: {
+    tailwindcss: {},
+    autoprefixer: {},
+  },
+};

tauri-app/src-tauri/Cargo.lock

@@ -47,6 +47,15 @@ version = "1.0.102"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7f202df86484c868dbad7eaa557ef785d5c66295e41b460ef922eca0723b842c"
 
+[[package]]
+name = "arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3d036a3c4ab069c7b410a2ce876bd74808d2d0888a82667669f8e783a898bf1"
+dependencies = [
+ "derive_arbitrary",
+]
+
 [[package]]
 name = "atk"
 version = "0.18.2"
@@ -521,6 +530,17 @@ dependencies = [
  "serde_core",
 ]
 
+[[package]]
+name = "derive_arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e567bd82dcff979e4b03460c307b3cdc9e96fde3d73bed1496d2bc75d9dd62a"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "derive_more"
 version = "0.99.20"
@@ -565,13 +585,34 @@ dependencies = [
  "crypto-common",
 ]
 
+[[package]]
+name = "dirs"
+version = "5.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "44c45a9d03d6676652bcb5e724c7e988de1acad23a711b5217ab9cbecbec2225"
+dependencies = [
+ "dirs-sys 0.4.1",
+]
+
 [[package]]
 name = "dirs"
 version = "6.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c3e8aa94d75141228480295a7d0e7feb620b1a5ad9f12bc40be62411e38cce4e"
 dependencies = [
- "dirs-sys",
+ "dirs-sys 0.5.0",
+]
+
+[[package]]
+name = "dirs-sys"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "520f05a5cbd335fae5a99ff7a6ab8627577660ee5cfd6a94a6a929b52ff0321c"
+dependencies = [
+ "libc",
+ "option-ext",
+ "redox_users 0.4.6",
+ "windows-sys 0.48.0",
 ]
 
 [[package]]
@@ -582,7 +623,7 @@ checksum = "e01a3366d27ee9890022452ee61b2b63a67e6f13f58900b651ff5665f0bb1fab"
 dependencies = [
  "libc",
  "option-ext",
- "redox_users",
+ "redox_users 0.5.2",
  "windows-sys 0.61.2",
 ]
 
@@ -830,12 +871,14 @@ name = "forgelm-chat"
 version = "0.1.0"
 dependencies = [
  "reqwest 0.12.28",
+ "runtime-deps",
  "serde",
  "serde_json",
  "tauri",
  "tauri-build",
  "tauri-plugin-dialog",
  "tauri-plugin-shell",
+ "tokio",
 ]
 
 [[package]]
@@ -1380,7 +1423,7 @@ dependencies = [
  "tokio",
  "tokio-rustls",
  "tower-service",
- "webpki-roots",
+ "webpki-roots 1.0.6",
 ]
 
 [[package]]
@@ -2787,6 +2830,17 @@ dependencies = [
  "bitflags 2.11.0",
 ]
 
+[[package]]
+name = "redox_users"
+version = "0.4.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba009ff324d1fc1b900bd1fdb31564febe58a8ccc8a6fdbb93b543d33b13ca43"
+dependencies = [
+ "getrandom 0.2.17",
+ "libredox",
+ "thiserror 1.0.69",
+]
+
 [[package]]
 name = "redox_users"
 version = "0.5.2"
@@ -2882,7 +2936,7 @@ dependencies = [
  "wasm-bindgen",
  "wasm-bindgen-futures",
  "web-sys",
- "webpki-roots",
+ "webpki-roots 1.0.6",
 ]
 
 [[package]]
@@ -2957,6 +3011,19 @@ dependencies = [
  "windows-sys 0.52.0",
 ]
 
+[[package]]
+name = "runtime-deps"
+version = "0.1.0"
+dependencies = [
+ "anyhow",
+ "dirs 5.0.1",
+ "serde",
+ "serde_json",
+ "sha2",
+ "ureq",
+ "zip",
+]
+
 [[package]]
 name = "rustc-hash"
 version = "2.1.1"
@@ -2978,6 +3045,7 @@ version = "0.23.37"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "758025cb5fccfd3bc2fd74708fd4682be41d99e5dff73c377c0646c6012c73a4"
 dependencies = [
+ "log",
  "once_cell",
  "ring",
  "rustls-pki-types",
@@ -3655,7 +3723,7 @@ dependencies = [
  "anyhow",
  "bytes",
  "cookie",
- "dirs",
+ "dirs 6.0.0",
  "dunce",
  "embed_plist",
  "getrandom 0.3.4",
@@ -3705,7 +3773,7 @@ checksum = "4bbc990d1dbf57a8e1c7fa2327f2a614d8b757805603c1b9ba5c81bade09fd4d"
 dependencies = [
  "anyhow",
  "cargo_toml",
- "dirs",
+ "dirs 6.0.0",
  "glob",
  "heck 0.5.0",
  "json-patch",
@@ -4268,7 +4336,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a5e85aa143ceb072062fc4d6356c1b520a51d636e7bc8e77ec94be3608e5e80c"
 dependencies = [
  "crossbeam-channel",
- "dirs",
+ "dirs 6.0.0",
  "libappindicator",
  "muda",
  "objc2",
@@ -4366,6 +4434,24 @@ version = "0.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8ecb6da28b8a351d773b68d5825ac39017e680750f980f3a1a85cd8dd28a47c1"
 
+[[package]]
+name = "ureq"
+version = "2.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "02d1a66277ed75f640d608235660df48c8e3c19f3b4edb6a263315626cc3c01d"
+dependencies = [
+ "base64 0.22.1",
+ "flate2",
+ "log",
+ "once_cell",
+ "rustls",
+ "rustls-pki-types",
+ "serde",
+ "serde_json",
+ "url",
+ "webpki-roots 0.26.11",
+]
+
 [[package]]
 name = "url"
 version = "2.5.8"
@@ -4678,6 +4764,15 @@ dependencies = [
  "system-deps",
 ]
 
+[[package]]
+name = "webpki-roots"
+version = "0.26.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "521bc38abb08001b01866da9f51eb7c5d647a19260e00054a8c7fd5f9e57f7a9"
+dependencies = [
+ "webpki-roots 1.0.6",
+]
+
 [[package]]
 name = "webpki-roots"
 version = "1.0.6"
@@ -4917,6 +5012,15 @@ dependencies = [
  "windows-targets 0.42.2",
 ]
 
+[[package]]
+name = "windows-sys"
+version = "0.48.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "677d2418bec65e3338edb076e806bc1ec15693c5d0104683f2efe857f61056a9"
+dependencies = [
+ "windows-targets 0.48.5",
+]
+
 [[package]]
 name = "windows-sys"
 version = "0.52.0"
@@ -4968,6 +5072,21 @@ dependencies = [
  "windows_x86_64_msvc 0.42.2",
 ]
 
+[[package]]
+name = "windows-targets"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9a2fa6e2155d7247be68c096456083145c183cbbbc2764150dda45a87197940c"
+dependencies = [
+ "windows_aarch64_gnullvm 0.48.5",
+ "windows_aarch64_msvc 0.48.5",
+ "windows_i686_gnu 0.48.5",
+ "windows_i686_msvc 0.48.5",
+ "windows_x86_64_gnu 0.48.5",
+ "windows_x86_64_gnullvm 0.48.5",
+ "windows_x86_64_msvc 0.48.5",
+]
+
 [[package]]
 name = "windows-targets"
 version = "0.52.6"
@@ -5025,6 +5144,12 @@ version = "0.42.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "597a5118570b68bc08d8d59125332c54f1ba9d9adeedeef5b99b02ba2b0698f8"
 
+[[package]]
+name = "windows_aarch64_gnullvm"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2b38e32f0abccf9987a4e3079dfb67dcd799fb61361e53e2882c3cbaf0d905d8"
+
 [[package]]
 name = "windows_aarch64_gnullvm"
 version = "0.52.6"
@@ -5043,6 +5168,12 @@ version = "0.42.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e08e8864a60f06ef0d0ff4ba04124db8b0fb3be5776a5cd47641e942e58c4d43"
 
+[[package]]
+name = "windows_aarch64_msvc"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dc35310971f3b2dbbf3f0690a219f40e2d9afcf64f9ab7cc1be722937c26b4bc"
+
 [[package]]
 name = "windows_aarch64_msvc"
 version = "0.52.6"
@@ -5061,6 +5192,12 @@ version = "0.42.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c61d927d8da41da96a81f029489353e68739737d3beca43145c8afec9a31a84f"
 
+[[package]]
+name = "windows_i686_gnu"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a75915e7def60c94dcef72200b9a8e58e5091744960da64ec734a6c6e9b3743e"
+
 [[package]]
 name = "windows_i686_gnu"
 version = "0.52.6"
@@ -5091,6 +5228,12 @@ version = "0.42.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "44d840b6ec649f480a41c8d80f9c65108b92d89345dd94027bfe06ac444d1060"
 
+[[package]]
+name = "windows_i686_msvc"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8f55c233f70c4b27f66c523580f78f1004e8b5a8b659e05a4eb49d4166cca406"
+
 [[package]]
 name = "windows_i686_msvc"
 version = "0.52.6"
@@ -5109,6 +5252,12 @@ version = "0.42.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8de912b8b8feb55c064867cf047dda097f92d51efad5b491dfb98f6bbb70cb36"
 
+[[package]]
+name = "windows_x86_64_gnu"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "53d40abd2583d23e4718fddf1ebec84dbff8381c07cae67ff7768bbf19c6718e"
+
 [[package]]
 name = "windows_x86_64_gnu"
 version = "0.52.6"
@@ -5127,6 +5276,12 @@ version = "0.42.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "26d41b46a36d453748aedef1486d5c7a85db22e56aff34643984ea85514e94a3"
 
+[[package]]
+name = "windows_x86_64_gnullvm"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b7b52767868a23d5bab768e390dc5f5c55825b6d30b86c844ff2dc7414044cc"
+
 [[package]]
 name = "windows_x86_64_gnullvm"
 version = "0.52.6"
@@ -5145,6 +5300,12 @@ version = "0.42.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9aec5da331524158c6d1a4ac0ab1541149c0b9505fde06423b02f5ef0106b9f0"
 
+[[package]]
+name = "windows_x86_64_msvc"
+version = "0.48.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ed94fce61571a4006852b7389a063ab983c02eb1bb37b47f8272ce92d06d9538"
+
 [[package]]
 name = "windows_x86_64_msvc"
 version = "0.52.6"
@@ -5295,7 +5456,7 @@ dependencies = [
  "block2",
  "cookie",
  "crossbeam-channel",
- "dirs",
+ "dirs 6.0.0",
  "dom_query",
  "dpi",
  "dunce",
@@ -5453,8 +5614,37 @@ dependencies = [
  "syn 2.0.117",
 ]
 
+[[package]]
+name = "zip"
+version = "2.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fabe6324e908f85a1c52063ce7aa26b68dcb7eb6dbc83a2d148403c9bc3eba50"
+dependencies = [
+ "arbitrary",
+ "crc32fast",
+ "crossbeam-utils",
+ "displaydoc",
+ "flate2",
+ "indexmap 2.13.0",
+ "memchr",
+ "thiserror 2.0.18",
+ "zopfli",
+]
+
 [[package]]
 name = "zmij"
 version = "1.0.21"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b8848ee67ecc8aedbaf3e4122217aff892639231befc6a1b58d29fff4c2cabaa"
+
+[[package]]
+name = "zopfli"
+version = "0.8.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f05cd8797d63865425ff89b5c4a48804f35ba0ce8d125800027ad6017d2b5249"
+dependencies = [
+ "bumpalo",
+ "crc32fast",
+ "log",
+ "simd-adler32",
+]

tauri-app/src-tauri/Cargo.toml

@@ -25,6 +25,8 @@ tauri-plugin-dialog = "2"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 reqwest = { version = "0.12", default-features = false, features = ["json", "rustls-tls"] }
+tokio = { version = "1", features = ["fs", "io-util", "sync", "time"] }
+runtime-deps = { path = "../../crates/runtime-deps" }
 
 [profile.release]
 panic = "abort"

tauri-app/src-tauri/src/lib.rs

@@ -1,3 +1,5 @@
+use std::sync::Arc;
+
 use tauri::{AppHandle, Manager, State};
 
 mod gpu;
@@ -9,9 +11,13 @@ use runtime::process::find_localllm_binary;
 use runtime::types::{RuntimeConfig, RuntimeSnapshot};
 use runtime::{bootstrap_runtime, get_runtime_snapshot, save_runtime_config_command, start_runtime, stop_runtime, RuntimeState};
 use gpu::query_gpu;
-use model_library::huggingface::search_hugging_face_models;
+use model_library::huggingface::{
+    cancel_hugging_face_download, search_hugging_face_models, start_hugging_face_download,
+    HuggingFaceDownloadState,
+};
 use model_library::local_index::list_entries;
-use model_library::paths::{managed_models_dir, model_library_index_path};
+use model_library::paths::{managed_models_dir, model_library_index_path, resolve_models_dir};
+use runtime::config::{load_runtime_config, save_runtime_config};
 use model_library::types::{ModelLibraryEntry, ModelLibraryState};
 use model_files::{import_model, pick_model_files, pick_models_dir, scan_models};
 
@@ -80,6 +86,105 @@ async fn search_hugging_face_models_command(
     search_hugging_face_models(&query, token.as_deref()).await
 }
 
+#[tauri::command]
+async fn start_hugging_face_download_command(
+    app: AppHandle,
+    state: State<'_, Arc<HuggingFaceDownloadState>>,
+    repo: String,
+    file_name: String,
+    token: Option<String>,
+) -> Result<String, String> {
+    // Load the user-configured models_dir (defaults to ./models if None/empty)
+    let config = load_runtime_config(&app)?;
+    let models_root = resolve_models_dir(config.models_dir.as_deref());
+
+    start_hugging_face_download(
+        app,
+        state.inner().clone(),
+        repo,
+        file_name,
+        token,
+        models_root,
+    )
+    .await
+}
+
+#[tauri::command]
+fn cancel_hugging_face_download_command(
+    state: State<'_, Arc<HuggingFaceDownloadState>>,
+    download_id: String,
+) -> Result<(), String> {
+    cancel_hugging_face_download(state.inner(), &download_id)
+}
+
+/// Swap the active model: stops the runtime, updates model_path, restarts.
+/// This is the hot-swap primitive — used when user picks a different local model.
+#[tauri::command]
+async fn swap_model_command(
+    app: AppHandle,
+    state: State<'_, RuntimeState>,
+    new_model_path: String,
+) -> Result<RuntimeSnapshot, String> {
+    // Stop first — this is a no-op if already stopped
+    stop_runtime(state.clone())?;
+
+    // Load, update model_path, save
+    let mut config = load_runtime_config(&app)?;
+    config.model_path = Some(new_model_path);
+    save_runtime_config(&app, &config)
+        .map_err(|e| format!("swap save failed: {e}"))?;
+
+    // Restart with the updated config
+    start_runtime(app, state)
+}
+
+/// Delete a local .gguf model file from disk.
+#[tauri::command]
+fn delete_model_file(file_path: String) -> Result<(), String> {
+    let path = std::path::Path::new(&file_path);
+    if !path.exists() {
+        return Err(format!("File not found: {file_path}"));
+    }
+    std::fs::remove_file(path).map_err(|e| format!("Delete failed: {e}"))
+}
+
+/// Open the folder containing a model file in the native file explorer
+/// and optionally select/highlight the file.
+#[tauri::command]
+fn open_model_folder_command(file_path: String) -> Result<(), String> {
+    let path = std::path::Path::new(&file_path);
+    let _folder = path.parent().unwrap_or(path);
+
+    #[cfg(target_os = "windows")]
+    {
+        std::process::Command::new("explorer")
+            .args(["/select,", &file_path])
+            .spawn()
+            .map_err(|e| format!("Failed to open Explorer: {e}"))?;
+    }
+    #[cfg(target_os = "macos")]
+    {
+        std::process::Command::new("open")
+            .args(["-R", &file_path])
+            .spawn()
+            .map_err(|e| format!("Failed to open Finder: {e}"))?;
+    }
+    #[cfg(target_os = "linux")]
+    {
+        std::process::Command::new("xdg-open")
+            .arg(folder)
+            .spawn()
+            .map_err(|e| format!("Failed to open file manager: {e}"))?;
+    }
+    #[cfg(not(any(target_os = "windows", target_os = "macos", target_os = "linux")))]
+    {
+        let _ = folder; // folder is only used on some platforms
+        return Err("Unsupported platform".into());
+    }
+
+    Ok(())
+}
+
 #[tauri::command]
 fn list_model_library_entries_command(app: AppHandle) -> Result<Vec<ModelLibraryEntry>, String> {
     let app_data_dir = app
@@ -112,6 +217,7 @@ pub fn run() {
         .plugin(tauri_plugin_shell::init())
         .plugin(tauri_plugin_dialog::init())
         .manage(RuntimeState::new())
+        .manage(Arc::new(HuggingFaceDownloadState::default()))
         .invoke_handler(tauri::generate_handler![
             query_gpu,
             scan_models,
@@ -119,6 +225,11 @@ pub fn run() {
             pick_model_files,
             import_model,
             search_hugging_face_models_command,
+            start_hugging_face_download_command,
+            cancel_hugging_face_download_command,
+            swap_model_command,
+            delete_model_file,
+            open_model_folder_command,
             list_model_library_entries_command,
             get_model_library_state_command,
             bootstrap_runtime_command,

tauri-app/src-tauri/src/model_files.rs

@@ -3,6 +3,8 @@ use std::path::Path;
 use tauri::AppHandle;
 use tauri_plugin_dialog::DialogExt;
 
+use crate::model_library::paths::resolve_models_dir;
+
 #[derive(serde::Serialize, serde::Deserialize, Clone, Debug, PartialEq, Eq)]
 pub struct ModelFile {
     pub name: String,
@@ -34,23 +36,64 @@ fn scan_model_dir(path: &Path) -> Result<Vec<ModelFile>, String> {
         return Err(format!("Not a directory: {}", path.display()));
     }
 
+    // Default scan is top-level only (no recursion into subdirectories)
+    scan_model_dir_top_level(path)
+}
+
+/// Scans only the immediate children of `path` (depth=1, no recursion).
+/// Returns only files with a `.gguf` extension (case-insensitive), sorted by name.
+fn scan_model_dir_top_level(path: &Path) -> Result<Vec<ModelFile>, String> {
+    if !path.exists() {
+        return Err(format!("Directory not found: {}", path.display()));
+    }
+    if !path.is_dir() {
+        return Err(format!("Not a directory: {}", path.display()));
+    }
+
     let entries = std::fs::read_dir(path).map_err(|e| e.to_string())?;
-    let mut models = Vec::new();
+    let mut models: Vec<ModelFile> = entries
+        .flatten()
+        .map(|e| e.path())
+        .filter(|p| {
+            p.is_file()
+                && p.extension()
+                    .map(|ext| ext.eq_ignore_ascii_case("gguf"))
+                    .unwrap_or(false)
+        })
+        .map(|p| model_file_from_path(&p))
+        .collect();
+
+    models.sort_by(|a, b| a.name.cmp(&b.name));
+    Ok(models)
+}
+
+/// Recursively scans all subdirectories for .gguf files.
+/// Kept for future use when deep scanning may be needed.
+#[allow(dead_code)]
+fn scan_model_dir_recursive(dir: &Path, models: &mut Vec<ModelFile>) -> Result<(), String> {
+    let entries = std::fs::read_dir(dir).map_err(|e| e.to_string())?;
 
     for entry in entries.flatten() {
         let path = entry.path();
-        if path.is_file()
+        if path.is_dir() {
+            // Skip hidden directories (e.g. .git, .cache)
+            if let Some(name) = path.file_name().and_then(|n| n.to_str()) {
+                if name.starts_with('.') {
+                    continue;
+                }
+            }
+            scan_model_dir_recursive(&path, models)?;
+        } else if path.is_file()
             && path
-            .extension()
-            .map(|ext| ext.eq_ignore_ascii_case("gguf"))
-            .unwrap_or(false)
+                .extension()
+                .map(|ext| ext.eq_ignore_ascii_case("gguf"))
+                .unwrap_or(false)
         {
             models.push(model_file_from_path(&path));
         }
     }
 
-    models.sort_by(|a, b| a.name.cmp(&b.name));
-    Ok(models)
+    Ok(())
 }
 
 fn import_model_file(src: &Path, dest_dir: &Path) -> Result<ModelFile, String> {
@@ -86,7 +129,12 @@ fn import_model_file(src: &Path, dest_dir: &Path) -> Result<ModelFile, String> {
 
 #[tauri::command]
 pub fn scan_models(dir: String) -> Result<Vec<ModelFile>, String> {
-    scan_model_dir(Path::new(&dir))
+    let path = if dir.trim().is_empty() {
+        resolve_models_dir(None)
+    } else {
+        resolve_models_dir(Some(&dir))
+    };
+    scan_model_dir_top_level(&path)
 }
 
 #[tauri::command]
@@ -171,6 +219,63 @@ mod tests {
         let _ = std::fs::remove_dir_all(&dir);
     }
 
+    #[test]
+    fn scan_model_dir_returns_only_top_level_gguf() {
+        // Tests that scan_model_dir (now using top-level scan) does NOT recurse into subdirectories
+        let dir = temp_path("scan-models-top-level");
+        let sub = dir.join("mradermacher").join("Suri-Qwen-3.5-4B-Uncensored-i1-GGUF");
+        std::fs::create_dir_all(&sub).expect("create nested dirs");
+        std::fs::write(dir.join("top-level.gguf"), b"demo").expect("write top-level model");
+        std::fs::write(sub.join("nested-model.gguf"), b"demo").expect("write nested model");
+        std::fs::write(sub.join("notes.txt"), b"ignore").expect("write non-model file");
+
+        let models = scan_model_dir(&dir).expect("scan model dir");
+        let names: Vec<_> = models.iter().map(|m| m.name.as_str()).collect();
+
+        // Only top-level model should be returned, NOT nested ones
+        assert_eq!(names, vec!["top-level.gguf"]);
+        assert!(models.iter().all(|m| m.path.ends_with(".gguf")));
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
+    #[test]
+    fn scan_model_dir_recursive_finds_nested_gguf() {
+        // Tests that the recursive scan function still works for future use
+        use super::scan_model_dir_recursive;
+        
+        let dir = temp_path("scan-models-recursive-fn");
+        let sub = dir.join("mradermacher").join("Suri-Qwen-3.5-4B-Uncensored-i1-GGUF");
+        std::fs::create_dir_all(&sub).expect("create nested dirs");
+        std::fs::write(dir.join("top-level.gguf"), b"demo").expect("write top-level model");
+        std::fs::write(sub.join("nested-model.gguf"), b"demo").expect("write nested model");
+
+        let mut models = Vec::new();
+        scan_model_dir_recursive(&dir, &mut models).expect("recursive scan");
+        models.sort_by(|a, b| a.name.cmp(&b.name));
+
+        let names: Vec<_> = models.iter().map(|m| m.name.as_str()).collect();
+        assert_eq!(names, vec!["nested-model.gguf", "top-level.gguf"]);
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
+    #[test]
+    fn scan_model_dir_skips_hidden_directories() {
+        let dir = temp_path("scan-models-hidden");
+        let hidden = dir.join(".git");
+        std::fs::create_dir_all(&hidden).expect("create hidden dir");
+        std::fs::write(dir.join("visible.gguf"), b"demo").expect("write visible model");
+        std::fs::write(hidden.join("secret.gguf"), b"demo").expect("write hidden model");
+
+        let models = scan_model_dir(&dir).expect("scan model dir");
+        let names: Vec<_> = models.into_iter().map(|m| m.name).collect();
+
+        assert_eq!(names, vec!["visible.gguf"]);
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
     #[test]
     fn import_model_file_returns_existing_model_when_source_matches_destination() {
         let dir = temp_path("import-model");

tauri-app/src-tauri/src/model_library/huggingface.rs

@@ -1,7 +1,69 @@
+use std::{
+    collections::HashMap,
+    path::{Path, PathBuf},
+    sync::{
+        atomic::{AtomicBool, Ordering},
+        Arc, Mutex,
+    },
+    time::{Duration, Instant},
+};
+
 use reqwest::header::{AUTHORIZATION, USER_AGENT};
 use serde::Deserialize;
+use tauri::{AppHandle, Emitter};
+use tokio::{fs, io::AsyncWriteExt};
 
-use super::types::{HuggingFaceModelResult, HuggingFaceVerifiedFile};
+use super::types::{
+    HuggingFaceDownloadPhase, HuggingFaceDownloadProgress, HuggingFaceModelResult,
+    HuggingFaceVerifiedFile,
+};
+
+pub const HUGGING_FACE_DOWNLOAD_EVENT: &str = "hugging-face-download-progress";
+pub const HUGGING_FACE_USER_AGENT: &str = "ForgeLM/0.1";
+
+#[derive(Default)]
+pub struct HuggingFaceDownloadState {
+    next_id: std::sync::atomic::AtomicU64,
+    cancellations: Mutex<HashMap<String, Arc<AtomicBool>>>,
+}
+
+impl HuggingFaceDownloadState {
+    pub fn start_download(&self) -> (String, Arc<AtomicBool>) {
+        let sequence = self
+            .next_id
+            .fetch_add(1, std::sync::atomic::Ordering::Relaxed)
+            + 1;
+        let download_id = format!("hf-download-{sequence}");
+        let cancellation = Arc::new(AtomicBool::new(false));
+        let mut guard = self
+            .cancellations
+            .lock()
+            .expect("hugging face download state poisoned");
+        guard.insert(download_id.clone(), cancellation.clone());
+        (download_id, cancellation)
+    }
+
+    pub fn cancel(&self, download_id: &str) -> bool {
+        let mut guard = self
+            .cancellations
+            .lock()
+            .expect("hugging face download state poisoned");
+        if let Some(flag) = guard.remove(download_id) {
+            flag.store(true, Ordering::Relaxed);
+            true
+        } else {
+            false
+        }
+    }
+
+    pub fn finish(&self, download_id: &str) {
+        let mut guard = self
+            .cancellations
+            .lock()
+            .expect("hugging face download state poisoned");
+        guard.remove(download_id);
+    }
+}
 
 #[derive(Deserialize, Debug)]
 struct HubSibling {
@@ -24,8 +86,359 @@ fn quantization_label(path: &str) -> Option<String> {
     stem.rsplit_once('-').map(|(_, suffix)| suffix.to_string())
 }
 
+fn validate_relative_segments(value: &str, label: &str) -> Result<Vec<String>, String> {
+    if value.trim().is_empty() {
+        return Err(format!("{label} cannot be empty"));
+    }
+    if value.contains('\\') {
+        return Err(format!("{label} contains invalid separators"));
+    }
+
+    let segments: Vec<String> = value.split('/').map(str::to_string).collect();
+    if segments
+        .iter()
+        .any(|segment| segment.is_empty() || segment == "." || segment == "..")
+    {
+        return Err(format!("{label} contains invalid path segments"));
+    }
+
+    Ok(segments)
+}
+
+fn managed_download_destination(
+    managed_root: &Path,
+    repo: &str,
+    file_name: &str,
+) -> Result<PathBuf, String> {
+    let repo_segments = validate_relative_segments(repo, "repo")?;
+    let file_segments = validate_relative_segments(file_name, "file name")?;
+
+    let mut destination = managed_root.to_path_buf();
+    for segment in repo_segments.into_iter().chain(file_segments.into_iter()) {
+        destination.push(segment);
+    }
+
+    Ok(destination)
+}
+
+pub fn hugging_face_resolve_url(repo: &str, file_name: &str) -> String {
+    format!("https://huggingface.co/{repo}/resolve/main/{file_name}")
+}
+
+pub fn download_progress_mib(bytes: u64) -> f64 {
+    bytes as f64 / 1_048_576.0
+}
+
+pub fn download_progress_percentage(downloaded_bytes: u64, total_bytes: Option<u64>) -> f64 {
+    match total_bytes {
+        Some(total) if total > 0 => {
+            ((downloaded_bytes as f64 / total as f64) * 100.0).clamp(0.0, 100.0)
+        }
+        _ => 0.0,
+    }
+}
+
+fn download_speed_mib_per_second(bytes_delta: u64, elapsed: Duration) -> f64 {
+    let seconds = elapsed.as_secs_f64();
+    if seconds <= 0.0 {
+        return 0.0;
+    }
+    download_progress_mib(bytes_delta) / seconds
+}
+
+fn emit_download_progress(app: &AppHandle, payload: &HuggingFaceDownloadProgress) -> Result<(), String> {
+    app.emit(HUGGING_FACE_DOWNLOAD_EVENT, payload)
+        .map_err(|error| format!("failed to emit download progress: {error}"))
+}
+
+fn build_progress_payload(
+    download_id: &str,
+    repo: &str,
+    file_name: &str,
+    phase: HuggingFaceDownloadPhase,
+    downloaded_bytes: u64,
+    total_bytes: Option<u64>,
+    download_speed: f64,
+    destination_path: Option<&Path>,
+    message: Option<String>,
+    error: Option<String>,
+) -> HuggingFaceDownloadProgress {
+    HuggingFaceDownloadProgress {
+        download_id: download_id.to_string(),
+        repo: repo.to_string(),
+        file_name: file_name.to_string(),
+        phase,
+        progress_percentage: download_progress_percentage(downloaded_bytes, total_bytes),
+        downloaded_mb: download_progress_mib(downloaded_bytes),
+        total_mb: total_bytes.map(download_progress_mib),
+        download_speed,
+        destination_path: destination_path.map(|path| path.to_string_lossy().to_string()),
+        message,
+        error,
+    }
+}
+
+async fn remove_if_exists(path: &Path) {
+    if fs::metadata(path).await.is_ok() {
+        let _ = fs::remove_file(path).await;
+    }
+}
+
+async fn run_hugging_face_download(
+    app: AppHandle,
+    state: Arc<HuggingFaceDownloadState>,
+    download_id: String,
+    repo: String,
+    file_name: String,
+    token: Option<String>,
+    cancellation: Arc<AtomicBool>,
+    managed_root: PathBuf,
+) -> Result<(), String> {
+    let destination = managed_download_destination(&managed_root, &repo, &file_name)?;
+    println!(
+        "[HF Download] download_id={} repo={} file={} destination={}",
+        download_id,
+        repo,
+        file_name,
+        destination.display()
+    );
+    let temp_path = destination.with_extension("part");
+    let parent = destination
+        .parent()
+        .ok_or_else(|| "failed to resolve download destination".to_string())?;
+
+    fs::create_dir_all(parent)
+        .await
+        .map_err(|error| format!("failed to prepare download directory: {error}"))?;
+
+    println!(
+        "[HF Download] download_id={} directory ready, initiating HTTP request to {}",
+        download_id,
+        hugging_face_resolve_url(&repo, &file_name)
+    );
+
+    let client = reqwest::Client::new();
+    let mut request = client
+        .get(hugging_face_resolve_url(&repo, &file_name))
+        .header(USER_AGENT, HUGGING_FACE_USER_AGENT);
+
+    if let Some(value) = token.as_deref().map(str::trim).filter(|value| !value.is_empty()) {
+        request = request.header(AUTHORIZATION, format!("Bearer {value}"));
+    }
+
+    let response = request
+        .send()
+        .await
+        .map_err(|error| format!("Hugging Face download request failed: {error}"))?
+        .error_for_status()
+        .map_err(|error| format!("Hugging Face download request failed: {error}"))?;
+
+    let total_bytes = response.content_length();
+
+    println!(
+        "[HF Download] download_id={} response received, content_length={:?}",
+        download_id,
+        total_bytes
+    );
+
+    emit_download_progress(
+        &app,
+        &build_progress_payload(
+            &download_id,
+            &repo,
+            &file_name,
+            HuggingFaceDownloadPhase::Started,
+            0,
+            total_bytes,
+            0.0,
+            Some(&destination),
+            Some("Download started".to_string()),
+            None,
+        ),
+    )?;
+
+    let mut stream = response;
+    let mut file = fs::File::create(&temp_path)
+        .await
+        .map_err(|error| format!("failed to create temp download file: {error}"))?;
+    let mut downloaded_bytes = 0_u64;
+    let mut last_emit = Instant::now();
+    let mut last_emit_bytes = 0_u64;
+
+    loop {
+        if cancellation.load(Ordering::Relaxed) {
+            println!(
+                "[HF Download] download_id={} CANCELLED at bytes={}/{:?}",
+                download_id, downloaded_bytes, total_bytes
+            );
+            let _ = file.flush().await;
+            drop(file);
+            remove_if_exists(&temp_path).await;
+            emit_download_progress(
+                &app,
+                &build_progress_payload(
+                    &download_id,
+                    &repo,
+                    &file_name,
+                    HuggingFaceDownloadPhase::Cancelled,
+                    downloaded_bytes,
+                    total_bytes,
+                    0.0,
+                    Some(&destination),
+                    Some("Download cancelled".to_string()),
+                    None,
+                ),
+            )?;
+            state.finish(&download_id);
+            return Ok(());
+        }
+
+        let next_chunk = stream
+            .chunk()
+            .await
+            .map_err(|error| format!("failed to read download stream: {error}"))?;
+
+        let Some(chunk) = next_chunk else {
+            break;
+        };
+
+        file.write_all(&chunk)
+            .await
+            .map_err(|error| format!("failed to write download chunk: {error}"))?;
+
+        downloaded_bytes += chunk.len() as u64;
+        let elapsed = last_emit.elapsed();
+        if elapsed >= Duration::from_millis(500) {
+            let speed = download_speed_mib_per_second(downloaded_bytes - last_emit_bytes, elapsed);
+            emit_download_progress(
+                &app,
+                &build_progress_payload(
+                    &download_id,
+                    &repo,
+                    &file_name,
+                    HuggingFaceDownloadPhase::Progress,
+                    downloaded_bytes,
+                    total_bytes,
+                    speed,
+                    Some(&destination),
+                    None,
+                    None,
+                ),
+            )?;
+            last_emit = Instant::now();
+            last_emit_bytes = downloaded_bytes;
+        }
+    }
+
+    file.flush()
+        .await
+        .map_err(|error| format!("failed to flush download file: {error}"))?;
+    drop(file);
+
+    remove_if_exists(&destination).await;
+    fs::rename(&temp_path, &destination)
+        .await
+        .map_err(|error| format!("failed to finalize download file: {error}"))?;
+
+    let elapsed = last_emit.elapsed();
+    let speed = download_speed_mib_per_second(downloaded_bytes - last_emit_bytes, elapsed);
+
+    println!(
+        "[HF Download] download_id={} COMPLETED downloaded_bytes={} destination={}",
+        download_id,
+        downloaded_bytes,
+        destination.display()
+    );
+
+    emit_download_progress(
+        &app,
+        &build_progress_payload(
+            &download_id,
+            &repo,
+            &file_name,
+            HuggingFaceDownloadPhase::Completed,
+            downloaded_bytes,
+            total_bytes,
+            speed,
+            Some(&destination),
+            Some("Download completed".to_string()),
+            None,
+        ),
+    )?;
+
+    state.finish(&download_id);
+    Ok(())
+}
+
+pub async fn start_hugging_face_download(
+    app: AppHandle,
+    state: Arc<HuggingFaceDownloadState>,
+    repo: String,
+    file_name: String,
+    token: Option<String>,
+    managed_root: PathBuf,
+) -> Result<String, String> {
+    let (download_id, cancellation) = state.start_download();
+    let task_app = app.clone();
+    let task_state = state.clone();
+    let task_download_id = download_id.clone();
+    let task_repo = repo.clone();
+    let task_file_name = file_name.clone();
+
+    tauri::async_runtime::spawn(async move {
+        if let Err(error) = run_hugging_face_download(
+            task_app.clone(),
+            task_state.clone(),
+            task_download_id.clone(),
+            task_repo.clone(),
+            task_file_name.clone(),
+            token,
+            cancellation,
+            managed_root,
+        )
+        .await
+        {
+            println!(
+                "[HF Download] download_id={} ERROR: {}",
+                task_download_id, error
+            );
+            task_state.finish(&task_download_id);
+            let _ = emit_download_progress(
+                &task_app,
+                &build_progress_payload(
+                    &task_download_id,
+                    &task_repo,
+                    &task_file_name,
+                    HuggingFaceDownloadPhase::Error,
+                    0,
+                    None,
+                    0.0,
+                    None,
+                    None,
+                    Some(error),
+                ),
+            );
+        }
+    });
+
+    Ok(download_id)
+}
+
+pub fn cancel_hugging_face_download(
+    state: &HuggingFaceDownloadState,
+    download_id: &str,
+) -> Result<(), String> {
+    if state.cancel(download_id) {
+        Ok(())
+    } else {
+        Err(format!("download {download_id} was not found"))
+    }
+}
+
 #[cfg(test)]
-pub fn filter_verified_gguf_files(payload: &str) -> Result<Vec<HuggingFaceVerifiedFile>, serde_json::Error> {
+pub fn filter_verified_gguf_files(
+    payload: &str,
+) -> Result<Vec<HuggingFaceVerifiedFile>, serde_json::Error> {
     let model: HubModel = serde_json::from_str(payload)?;
     Ok(verified_files_from_model(model))
 }
@@ -66,7 +479,7 @@ pub async fn search_hugging_face_models(
             ("limit", "12"),
             ("full", "true"),
         ])
-        .header(USER_AGENT, "ForgeLM/0.1");
+        .header(USER_AGENT, HUGGING_FACE_USER_AGENT);
 
     if let Some(value) = token.map(str::trim).filter(|value| !value.is_empty()) {
         request = request.header(AUTHORIZATION, format!("Bearer {value}"));
@@ -109,7 +522,12 @@ pub async fn search_hugging_face_models(
 
 #[cfg(test)]
 mod tests {
-    use super::filter_verified_gguf_files;
+    use std::path::Path;
+
+    use super::{
+        download_progress_mib, download_progress_percentage, filter_verified_gguf_files,
+        hugging_face_resolve_url, managed_download_destination,
+    };
 
     #[test]
     fn search_hf_gguf_filters_verified_files_and_excludes_mmproj() {
@@ -132,4 +550,52 @@ mod tests {
         assert_eq!(files[1].path, "subdir/model-q8_0.gguf");
         assert_eq!(files[1].quantization.as_deref(), Some("q8_0"));
     }
+
+    #[test]
+    fn builds_direct_hugging_face_resolve_urls() {
+        assert_eq!(
+            hugging_face_resolve_url("microsoft/phi-4-mini", "phi-4-q4.gguf"),
+            "https://huggingface.co/microsoft/phi-4-mini/resolve/main/phi-4-q4.gguf"
+        );
+    }
+
+    #[test]
+    fn converts_bytes_to_mebibytes_for_progress_reporting() {
+        assert_eq!(download_progress_mib(1_048_576), 1.0);
+        assert_eq!(download_progress_mib(2_621_440), 2.5);
+    }
+
+    #[test]
+    fn calculates_progress_percentage_from_content_length() {
+        assert_eq!(download_progress_percentage(524_288, Some(1_048_576)), 50.0);
+        assert_eq!(download_progress_percentage(1_048_576, Some(1_048_576)), 100.0);
+        assert_eq!(download_progress_percentage(1_048_576, None), 0.0);
+    }
+
+    #[test]
+    fn builds_managed_destination_from_repo_and_file_path() {
+        let destination = managed_download_destination(
+            Path::new("/models"),
+            "microsoft/phi-4",
+            "quantized/phi-4-q4.gguf",
+        )
+        .expect("destination path");
+
+        assert_eq!(
+            destination,
+            Path::new("/models")
+                .join("microsoft")
+                .join("phi-4")
+                .join("quantized")
+                .join("phi-4-q4.gguf")
+        );
+    }
+
+    #[test]
+    fn rejects_parent_directory_segments_in_download_paths() {
+        let error = managed_download_destination(Path::new("/models"), "microsoft/../phi-4", "model.gguf")
+            .expect_err("invalid path should fail");
+
+        assert!(error.contains("invalid path segments"));
+    }
 }

tauri-app/src-tauri/src/model_library/local_index.rs

@@ -22,6 +22,7 @@ pub fn load_index(path: &Path) -> Result<ModelLibraryIndex, String> {
     }
 }
 
+#[allow(dead_code)]
 pub fn save_index(path: &Path, index: &ModelLibraryIndex) -> Result<(), String> {
     if let Some(parent) = path.parent() {
         fs::create_dir_all(parent)

tauri-app/src-tauri/src/model_library/paths.rs

@@ -3,6 +3,18 @@ use std::path::{Path, PathBuf};
 const MODEL_LIBRARY_DIR: &str = "model-library";
 const INDEX_FILE: &str = "index.json";
 const MANAGED_MODELS_DIR: &str = "managed-models";
+const DEFAULT_MODELS_DIR: &str = "models";
+
+/// Resolves the models directory from the user-configured `models_dir`.
+/// - `None`        → `./models` (relative to the app's working directory)
+/// - `Some("")`    → `./models`
+/// - `Some(path)`  → that exact path
+pub fn resolve_models_dir(models_dir: Option<&str>) -> PathBuf {
+    match models_dir {
+        None | Some("") => PathBuf::from(DEFAULT_MODELS_DIR),
+        Some(path) => PathBuf::from(path),
+    }
+}
 
 pub fn model_library_dir(app_data_dir: &Path) -> PathBuf {
     app_data_dir.join(MODEL_LIBRARY_DIR)
@@ -12,6 +24,9 @@ pub fn model_library_index_path(app_data_dir: &Path) -> PathBuf {
     model_library_dir(app_data_dir).join(INDEX_FILE)
 }
 
+/// Legacy path for the managed-models location inside AppData.
+/// Used only for index persistence — downloads now go to the user-configured
+/// `models_dir` (defaults to `./models` in the project root).
 pub fn managed_models_dir(app_data_dir: &Path) -> PathBuf {
     model_library_dir(app_data_dir).join(MANAGED_MODELS_DIR)
 }

tauri-app/src-tauri/src/model_library/types.rs

@@ -46,6 +46,32 @@ pub struct HuggingFaceModelResult {
     pub verified_files: Vec<HuggingFaceVerifiedFile>,
 }
 
+#[derive(Serialize, Deserialize, Clone, Debug, PartialEq, Eq)]
+#[serde(rename_all = "snake_case")]
+pub enum HuggingFaceDownloadPhase {
+    Started,
+    Progress,
+    Completed,
+    Cancelled,
+    Error,
+}
+
+#[derive(Serialize, Deserialize, Clone, Debug, PartialEq)]
+#[serde(rename_all = "camelCase")]
+pub struct HuggingFaceDownloadProgress {
+    pub download_id: String,
+    pub repo: String,
+    pub file_name: String,
+    pub phase: HuggingFaceDownloadPhase,
+    pub progress_percentage: f64,
+    pub downloaded_mb: f64,
+    pub total_mb: Option<f64>,
+    pub download_speed: f64,
+    pub destination_path: Option<String>,
+    pub message: Option<String>,
+    pub error: Option<String>,
+}
+
 #[derive(Serialize, Deserialize, Clone, Debug, PartialEq, Eq, Default)]
 #[serde(default)]
 pub struct ModelLibraryState {

tauri-app/src-tauri/src/runtime/mod.rs

@@ -118,7 +118,7 @@ fn start_runtime_inner(
         inner.config.clone()
     };
 
-    let child = match spawn_localllm(&config) {
+    let child = match spawn_localllm(app, &config) {
         Ok(child) => child,
         Err(message) => {
             let mut inner = state.0.lock().map_err(|e| e.to_string())?;

tauri-app/src-tauri/src/runtime/process.rs

@@ -1,7 +1,13 @@
 use std::path::PathBuf;
 use std::process::{Child, Command};
 
+use tauri::{AppHandle, Manager};
+
 use super::types::RuntimeConfig;
+use runtime_deps::{
+    apply_runtime_to_command, ensure_llama_runtime, recommended_variant_override_hint,
+    EnsureRuntimeRequest,
+};
 
 pub fn find_localllm_binary() -> Option<PathBuf> {
     let exe_name = if cfg!(target_os = "windows") {
@@ -52,7 +58,17 @@ pub fn find_localllm_binary() -> Option<PathBuf> {
     None
 }
 
-pub fn spawn_localllm(config: &RuntimeConfig) -> Result<Child, String> {
+fn runtime_cache_root(app: &AppHandle) -> Result<PathBuf, String> {
+    let app_data_dir = app
+        .path()
+        .app_data_dir()
+        .map_err(|error| format!("Cannot resolve app data dir: {error}"))?;
+    std::fs::create_dir_all(&app_data_dir)
+        .map_err(|error| format!("Cannot create app data dir: {error}"))?;
+    Ok(app_data_dir.join("runtime-cache"))
+}
+
+pub fn spawn_localllm(app: &AppHandle, config: &RuntimeConfig) -> Result<Child, String> {
     let model_path = config
         .model_path
         .as_deref()
@@ -65,11 +81,28 @@ pub fn spawn_localllm(config: &RuntimeConfig) -> Result<Child, String> {
             .to_string()
     })?;
 
-    Command::new(&binary_path)
+    let runtime = ensure_llama_runtime(&EnsureRuntimeRequest {
+        cache_root: &runtime_cache_root(app)?,
+        gpu_layers: config.gpu_layers,
+        variant_override: None,
+        binary_path: Some(&binary_path),
+    })
+    .map_err(|error| {
+        format!(
+            "Failed to prepare llama runtime dependencies: {error}\n{}",
+            recommended_variant_override_hint()
+        )
+    })?;
+
+    let mut command = Command::new(&binary_path);
+    command
         .env("LOCALLLM_MODEL", model_path)
         .env("LOCALLLM_ADDR", &config.addr)
         .env("LOCALLLM_GPU_LAYERS", config.gpu_layers.to_string())
-        .env("LOCALLLM_CTX", config.ctx_size.to_string())
+        .env("LOCALLLM_CTX", config.ctx_size.to_string());
+    apply_runtime_to_command(&mut command, &runtime);
+
+    command
         .spawn()
         .map_err(|e| format!("Failed to start server ({}): {e}", binary_path.display()))
 }

tauri-app/src-tauri/src/runtime/types.rs

@@ -2,6 +2,8 @@ use serde::{Deserialize, Serialize};
 
 use crate::model_library::types::RemotePreset;
 
+pub const DEFAULT_CTX_SIZE: u32 = 4096;
+
 #[derive(Serialize, Deserialize, Clone, Copy, PartialEq, Eq, Default, Debug)]
 #[serde(rename_all = "snake_case")]
 pub enum ConnectionMode {
@@ -102,7 +104,7 @@ impl Default for RuntimeConfig {
             models_dir: None,
             addr: Self::default_addr(),
             gpu_layers: 0,
-            ctx_size: 4096,
+            ctx_size: DEFAULT_CTX_SIZE,
             extra_model_paths: Vec::new(),
             selected_model_id: None,
             remote_presets: Vec::new(),
@@ -113,7 +115,14 @@ impl Default for RuntimeConfig {
 
 #[cfg(test)]
 mod tests {
-    use super::{ConnectionMode, RuntimeConfig};
+    use super::{ConnectionMode, RuntimeConfig, DEFAULT_CTX_SIZE};
+
+    #[test]
+    fn runtime_config_defaults_to_safe_context_window() {
+        let config = RuntimeConfig::default();
+
+        assert_eq!(config.ctx_size, DEFAULT_CTX_SIZE);
+    }
 
     #[test]
     fn runtime_config_deserializes_old_shape_with_new_defaults() {

tauri-app/src/App.tsx

@@ -1,11 +1,15 @@
-import { useState, useEffect, useCallback } from "react";
+import { useState, useEffect, useCallback, useMemo } from "react";
 import Sidebar from "./components/Sidebar";
 import ChatView from "./components/ChatView";
+import SettingsPage from "./features/settings/SettingsPage";
+import { ToastProvider } from "./components/Toast";
 import { useDB } from "./hooks/useDB";
 import { useModels } from "./hooks/useModels";
 import { useRuntimeManager } from "./features/runtime/useRuntimeManager";
 import { useRuntimeReadyModelRefresh } from "./features/runtime/useRuntimeReadyModelRefresh";
+import { swapModel } from "./features/runtime/api";
 import type { Chat } from "./lib/types";
+import type { RuntimeModelFile } from "./features/runtime/types";
 import "./App.css";
 
 export default function App() {
@@ -13,6 +17,7 @@ export default function App() {
   const [currentChatId, setCurrentChatId] = useState<number | null>(null);
   const [chats, setChats] = useState<Chat[]>([]);
   const [sidebarOpen, setSidebarOpen] = useState(false);
+  const [view, setView] = useState<"chat" | "settings">("chat");
 
   const runtime = useRuntimeManager();
   const { models, selectedModel, setSelectedModel, status, refresh: refreshModels } = useModels(
@@ -55,32 +60,51 @@ export default function App() {
     setSidebarOpen(false);
   }, []);
 
+  const [isSwapping, setIsSwapping] = useState(false);
+
+  const handleModelSwap = useCallback(async (newModelPath: string) => {
+    setIsSwapping(true);
+    try {
+      await swapModel(newModelPath);
+      setSelectedModel(newModelPath);
+    } catch (err) {
+      console.error("[App] handleModelSwap failed:", err);
+    } finally {
+      setIsSwapping(false);
+    }
+  }, [setSelectedModel]);
+
+  const localModels: RuntimeModelFile[] = useMemo(() => {
+    return runtime.scannedModels ?? [];
+  }, [runtime.scannedModels]);
+
   if (!ready) {
     return (
-      <div className="app-loading">
-        <div className="loading-spinner" />
-        <span>Initializing…</span>
-      </div>
+      <ToastProvider>
+        <div className="app-loading">
+          <div className="loading-spinner" />
+          <span>Initializing…</span>
+        </div>
+      </ToastProvider>
     );
   }
 
   return (
-    <div className={`app-layout${sidebarOpen ? " sidebar-open" : ""}`}>
-      <Sidebar
-        chats={chats}
-        currentChatId={currentChatId}
-        onSelectChat={handleSelectChat}
-        onNewChat={newChat}
-        onDeleteChat={deleteChat}
-        models={models}
-        selectedModel={selectedModel}
-        onModelChange={setSelectedModel}
-        status={status}
-        onRefreshModels={refreshModels}
-        runtime={runtime}
-        onCloseMobile={() => setSidebarOpen(false)}
-      />
-      {sidebarOpen && (
+    <ToastProvider>
+    <div className={`app-layout${sidebarOpen && view !== "settings" ? " sidebar-open" : ""}`}>
+      {view !== "settings" && (
+        <Sidebar
+          chats={chats}
+          currentChatId={currentChatId}
+          initialScreen={currentChatId === null}
+          onSelectChat={handleSelectChat}
+          onNewChat={newChat}
+          onDeleteChat={deleteChat}
+          onOpenSettings={() => setView("settings")}
+          onCloseMobile={() => setSidebarOpen(false)}
+        />
+      )}
+      {view !== "settings" && sidebarOpen && (
         <button
           type="button"
           className="app-sidebar-backdrop"
@@ -89,17 +113,29 @@ export default function App() {
         />
       )}
       <div className="app-main">
-        <ChatView
-          db={db!}
-          chatId={currentChatId}
-          server={runtime.connection.serverUrl}
-          apiKey={runtime.connection.apiKey}
-          model={selectedModel}
-          onChatCreated={(id) => { setCurrentChatId(id); loadChats(); setSidebarOpen(false); }}
-          onChatUpdated={loadChats}
-          onOpenSidebar={() => setSidebarOpen(true)}
-        />
+        {view === "settings" ? (
+          <SettingsPage runtime={runtime} onClose={() => setView("chat")} onModelSwapped={setSelectedModel} />
+        ) : (
+          <ChatView
+            db={db!}
+            chatId={currentChatId}
+            server={runtime.connection.serverUrl}
+            apiKey={runtime.connection.apiKey}
+            model={selectedModel}
+            models={models}
+            localModels={localModels}
+            selectedModel={selectedModel}
+            onModelChange={setSelectedModel}
+            onModelSwap={handleModelSwap}
+            isSwapping={isSwapping}
+            status={status}
+            onChatCreated={(id) => { setCurrentChatId(id); loadChats(); setSidebarOpen(false); }}
+            onChatUpdated={loadChats}
+            onOpenSidebar={() => setSidebarOpen(true)}
+          />
+        )}
       </div>
     </div>
+    </ToastProvider>
   );
 }

tauri-app/src/components/ChatView.test.tsx

@@ -17,6 +17,15 @@ function createDbStub(): DB {
   } as unknown as DB;
 }
 
+// Default props for ChatView tests
+const defaultChatViewProps = {
+  models: [] as import("../lib/types").ModelInfo[],
+  localModels: [] as import("../features/runtime/types").RuntimeModelFile[],
+  selectedModel: "",
+  onModelChange: vi.fn(),
+  status: "error" as import("../lib/types").ServerStatus,
+};
+
 describe("ChatView", () => {
   let container: HTMLDivElement;
   let root: Root;
@@ -51,6 +60,7 @@ describe("ChatView", () => {
           model=""
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
     });
@@ -68,7 +78,7 @@ describe("ChatView", () => {
     });
 
     expect(textarea?.value).toBe("Keep this draft");
-    expect(container.textContent).toContain("Connect an API");
+    expect(container.textContent).toContain("Runtime offline");
     expect(vi.mocked(db.run)).not.toHaveBeenCalled();
   });
 
@@ -103,6 +113,7 @@ describe("ChatView", () => {
           model="demo"
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
     });
@@ -119,6 +130,7 @@ describe("ChatView", () => {
           model="demo"
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
     });
@@ -192,6 +204,7 @@ describe("ChatView", () => {
           model="demo"
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
     });
@@ -221,6 +234,7 @@ describe("ChatView", () => {
           model="demo"
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
     });
@@ -310,6 +324,7 @@ describe("ChatView", () => {
           model="demo"
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
     });
@@ -382,6 +397,7 @@ describe("ChatView", () => {
           model="demo"
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
     });
@@ -420,6 +436,7 @@ describe("ChatView", () => {
           model="demo"
           onChatCreated={vi.fn()}
           onChatUpdated={vi.fn()}
+          {...defaultChatViewProps}
         />,
       );
       await Promise.resolve();

tauri-app/src/components/MessageBubble.css

@@ -1,7 +1,7 @@
-/* Edit mode */
+/* ── Edit mode ── */
 .msg-edit-container {
   width: 100%;
-  max-width: 780px;
+  max-width: 800px;
 }
 
 .msg-edit-textarea {
@@ -12,17 +12,19 @@
   color: var(--text);
   font-size: 0.9rem;
   font-family: inherit;
-  padding: 10px 14px;
+  padding: 11px 14px;
   resize: none;
   min-height: 80px;
   outline: none;
   line-height: 1.6;
+  box-shadow: 0 0 0 3px rgba(72,166,255,0.1);
 }
 
 .msg-edit-actions {
   display: flex;
   gap: 6px;
   margin-top: 8px;
+  flex-wrap: wrap;
 }
 
 .btn-primary {
@@ -31,13 +33,18 @@
   border: none;
   border-radius: var(--radius-sm);
   padding: 6px 14px;
+  min-height: 44px;
   font-size: 0.8rem;
   font-weight: 600;
   cursor: pointer;
   font-family: inherit;
-  transition: background 0.15s;
+  transition: background 0.15s, box-shadow 0.15s, transform 0.15s;
 }
 .btn-primary:hover { background: var(--accent2); }
+.btn-primary:focus-visible {
+  outline: none;
+  box-shadow: 0 0 0 3px rgba(72, 166, 255, 0.22);
+}
 
 .btn-secondary {
   background: var(--bg4);
@@ -45,108 +52,286 @@
   border: 1px solid var(--border);
   border-radius: var(--radius-sm);
   padding: 6px 14px;
+  min-height: 44px;
   font-size: 0.8rem;
   cursor: pointer;
   font-family: inherit;
-  transition: background 0.15s;
+  transition: background 0.15s, border-color 0.15s, color 0.15s, box-shadow 0.15s;
 }
 .btn-secondary:hover { background: var(--bg5); color: var(--text); }
+.btn-secondary:focus-visible {
+  outline: none;
+  box-shadow: 0 0 0 3px rgba(72, 166, 255, 0.18);
+}
 
 .btn-ghost {
-  background: none;
-  color: var(--text3);
-  border: none;
+  background: transparent;
+  color: var(--text2);
+  border: 1px solid transparent;
   border-radius: var(--radius-sm);
   padding: 6px 10px;
+  min-height: 44px;
   font-size: 0.8rem;
   cursor: pointer;
   font-family: inherit;
-  transition: color 0.15s;
+  transition: color 0.15s, background 0.15s, border-color 0.15s;
 }
-.btn-ghost:hover { color: var(--text2); }
-
-/* Message actions */
-.reasoning-panel {
-  width: 100%;
-  margin-bottom: 8px;
+.btn-ghost:hover { color: var(--text); background: rgba(237, 245, 255, 0.05); border-color: rgba(93, 114, 136, 0.28); }
+.btn-ghost:focus-visible {
+  outline: none;
+  box-shadow: 0 0 0 3px rgba(72, 166, 255, 0.18);
 }
 
-.reasoning-toggle {
-  display: inline-flex;
+/* ── Thinking inside bubble ── */
+.msg-thinking {
+  border-bottom: 1px solid rgba(255, 255, 255, 0.05);
+  background: transparent;
+}
+
+.msg-thinking > summary {
+  list-style: none;
+}
+
+.msg-thinking > summary::-webkit-details-marker {
+  display: none;
+}
+
+.thinking-summary {
+  display: flex;
   align-items: center;
   gap: 8px;
-  background: rgba(124, 158, 248, 0.08);
-  color: var(--text2);
-  border: 1px solid var(--border);
-  border-radius: 999px;
-  padding: 6px 12px;
-  font-size: 0.75rem;
+  padding: 10px 16px;
   cursor: pointer;
-  font-family: inherit;
+  font-size: 0.75rem;
+  font-weight: 500;
+  text-transform: uppercase;
+  letter-spacing: 0.05em;
+  color: rgba(255, 255, 255, 0.5);
+  user-select: none;
+  transition: color 0.2s ease-out;
 }
 
-.reasoning-toggle:hover {
-  background: rgba(124, 158, 248, 0.12);
-  color: var(--text);
+.thinking-summary:hover {
+  color: rgba(255, 255, 255, 0.8);
 }
 
-.reasoning-content {
-  margin-top: 8px;
-  padding: 12px 14px;
-  background: rgba(124, 158, 248, 0.06);
-  border: 1px solid var(--border);
-  border-radius: var(--radius);
-  color: var(--text2);
-  font-size: 0.84rem;
+.thinking-summary:focus-visible {
+  outline: none;
+  box-shadow: inset 0 0 0 2px rgba(72, 166, 255, 0.3);
+}
+
+.thinking-chevron {
+  width: 14px;
+  height: 14px;
+  stroke: currentColor;
+  stroke-width: 1.8;
+  fill: none;
+  stroke-linecap: round;
+  stroke-linejoin: round;
+  transition: transform 0.2s ease-out;
+  flex-shrink: 0;
+}
+
+.msg-thinking[open] .thinking-chevron {
+  transform: rotate(90deg);
+}
+
+.thinking-content {
+  padding: 4px 16px 16px 16px;
+  font-size: 0.875rem;
+  color: rgba(255, 255, 255, 0.6);
   line-height: 1.6;
+  font-style: normal;
 }
 
+.thinking-content > :first-child {
+  border-left: 2px solid rgba(255, 255, 255, 0.1);
+  margin-left: 16px;
+  padding-left: 12px;
+}
+
+/* ── Message bubble layout ── */
+.msg-bubble {
+  display: flex;
+  flex-direction: column;
+  padding: 0;
+  overflow: hidden;
+}
+
+.msg-content {
+  padding: 13px 16px;
+}
+
+/* ── Message metrics ── */
 .message-metrics {
   display: flex;
-  flex-wrap: wrap;
+  align-items: center;
   gap: 8px;
-  margin-top: 8px;
-  color: var(--text3);
-  font-size: 0.72rem;
-  font-variant-numeric: tabular-nums;
+  padding-left: 12px;
+  font-size: 0.75rem;
+  opacity: 0.7;
+  font-family: monospace;
+  color: #a8b8cc;
 }
 
-.message-metrics span {
-  background: var(--bg3);
-  border: 1px solid var(--border);
-  border-radius: 999px;
-  padding: 3px 8px;
+/* ── Message footer ── */
+.msg-footer {
+  display: flex;
+  justify-content: space-between;
+  align-items: stretch;
+  border-top: 1px solid rgba(255, 255, 255, 0.08);
+  background: rgba(0, 0, 0, 0.12);
+  min-height: 32px;
 }
 
+/* ── Compact actions for user bubble (floating bar) ── */
+.msg-wrapper.user {
+  position: relative;
+  margin-bottom: 24px;
+}
+
+.compact-actions {
+  position: absolute;
+  bottom: -12px;
+  right: 0;
+  display: flex;
+  gap: 4px;
+  padding: 4px;
+  border-radius: 8px;
+  background: var(--bg2);
+  border: 1px solid rgba(255, 255, 255, 0.1);
+  box-shadow: 0 4px 12px rgba(0, 0, 0, 0.2);
+  opacity: 0;
+  pointer-events: none;
+  transform: translateY(-4px);
+  transition: all 0.2s ease-out;
+  z-index: 10;
+}
+
+.msg-wrapper.user:hover .compact-actions,
+.msg-wrapper.user:focus-within .compact-actions {
+  opacity: 1;
+  pointer-events: auto;
+  transform: translateY(0);
+}
+
+.compact-actions .action-btn {
+  min-height: 28px;
+  min-width: 28px;
+  width: 28px;
+  height: 28px;
+  padding: 0;
+  background: transparent;
+  border: none;
+  border-radius: 4px;
+  color: #c8d7ea;
+}
+.compact-actions .action-btn svg {
+  width: 14px;
+  height: 14px;
+}
+.compact-actions .action-btn:hover {
+  background: rgba(255, 255, 255, 0.08);
+  color: var(--text);
+}
+.compact-actions .action-btn:focus-visible {
+  outline: none;
+  box-shadow: 0 0 0 2px rgba(72, 166, 255, 0.4);
+  opacity: 1;
+}
+.compact-actions .action-btn.copied {
+  color: var(--success);
+  background: rgba(48, 209, 88, 0.12);
+}
+.compact-actions .action-btn.danger {
+  color: #f4c2c2;
+}
+.compact-actions .action-btn.danger:hover {
+  background: rgba(240, 64, 64, 0.12);
+  color: #ffd7d7;
+}
+
+/* ── Message actions (assistant) ── */
 .msg-actions {
   display: flex;
-  flex-wrap: wrap;
-  gap: 4px;
-  margin-top: 5px;
   opacity: 0;
-  transition: opacity 0.15s;
+  transition: opacity 0.2s ease-out;
+}
+
+.msg-bubble:hover .msg-actions,
+.msg-bubble:focus-within .msg-actions {
+  opacity: 1;
 }
-.msg-wrapper:hover .msg-actions { opacity: 1; }
-.msg-wrapper:focus-within .msg-actions { opacity: 1; }
 
 .action-btn {
-  background: none;
-  border: 1px solid var(--border);
-  color: var(--text2);
-  border-radius: var(--radius-sm);
-  padding: 3px 9px;
-  font-size: 0.72rem;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  min-height: 32px;
+  min-width: 36px;
+  background: transparent;
+  border: none;
+  border-left: 1px solid rgba(255, 255, 255, 0.08);
+  color: #c8d7ea;
+  border-radius: 0;
+  padding: 0 12px;
   cursor: pointer;
-  transition: all 0.15s;
+  transition: background 0.2s ease-out, color 0.2s ease-out;
   font-family: inherit;
-  white-space: nowrap;
 }
-.action-btn:hover { background: var(--bg3); color: var(--text); border-color: var(--border2); }
-.action-btn.copied { color: var(--success); border-color: var(--success); }
-.action-btn.danger:hover { background: rgba(240,64,64,0.1); color: var(--danger); border-color: var(--danger); }
+.action-btn:first-child {
+  border-left: none;
+}
+.action-btn svg {
+  width: 16px;
+  height: 16px;
+  stroke: currentColor;
+  stroke-width: 1.5;
+  fill: none;
+  stroke-linecap: round;
+  stroke-linejoin: round;
+}
+.action-btn:hover {
+  background: rgba(255, 255, 255, 0.06);
+  color: var(--text);
+}
+.action-btn:focus-visible {
+  outline: none;
+  box-shadow: inset 0 0 0 2px rgba(72, 166, 255, 0.3);
+  opacity: 1;
+}
+.action-btn.copied  {
+  color: var(--success);
+  background: rgba(48, 209, 88, 0.08);
+}
+.action-btn.danger {
+  color: #f4c2c2;
+}
+.action-btn.danger:hover {
+  background: rgba(240, 64, 64, 0.1);
+  color: #ffd7d7;
+}
 
 @media (hover: none), (max-width: 960px) {
+  .msg-actions { opacity: 1; }
+}
+
+@media (max-width: 640px) {
   .msg-actions {
-    opacity: 1;
+    width: 100%;
+  }
+
+  .action-btn {
+    flex: 1 1 132px;
+  }
+}
+
+@media (prefers-reduced-motion: reduce) {
+  .action-btn,
+  .btn-primary,
+  .btn-secondary,
+  .btn-ghost {
+    animation: none;
+    transition: none;
   }
 }

tauri-app/src/components/MessageBubble.test.tsx

@@ -0,0 +1,72 @@
+/* @vitest-environment jsdom */
+
+import { createRoot, type Root } from "react-dom/client";
+import { act } from "react-dom/test-utils";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+import MessageBubble from "./MessageBubble";
+import type { Message } from "../lib/types";
+
+describe("MessageBubble", () => {
+  let container: HTMLDivElement;
+  let root: Root;
+
+  beforeEach(() => {
+    // @ts-expect-error React test environment flag
+    globalThis.IS_REACT_ACT_ENVIRONMENT = true;
+    container = document.createElement("div");
+    document.body.appendChild(container);
+    root = createRoot(container);
+    Object.assign(navigator, {
+      clipboard: {
+        writeText: vi.fn(() => Promise.resolve()),
+      },
+    });
+  });
+
+  afterEach(() => {
+    act(() => root.unmount());
+    container.remove();
+    vi.restoreAllMocks();
+    // @ts-expect-error React test environment flag
+    globalThis.IS_REACT_ACT_ENVIRONMENT = false;
+  });
+
+  it("renders accessible thinking and action controls for assistant messages", async () => {
+    const message: Message = {
+      id: 12,
+      chat_id: 4,
+      role: "assistant",
+      content: "Final answer",
+      reasoning_content: "Reasoning details",
+      prompt_tokens: 18,
+      completion_tokens: 24,
+      total_tokens: 42,
+      tokens_per_second: 6.4,
+      created_at: Date.now(),
+    };
+
+    await act(async () => {
+      root.render(
+        <MessageBubble
+          message={message}
+          onDelete={vi.fn()}
+          onEditAssistant={vi.fn()}
+          renderMarkdown={(text) => `<p>${text}</p>`}
+        />,
+      );
+    });
+
+    const summary = container.querySelector("summary.reasoning-toggle");
+    expect(summary?.getAttribute("aria-label")).toBe("Toggle thinking details");
+
+    const buttons = Array.from(container.querySelectorAll("button"));
+    expect(buttons.some((node) => node.getAttribute("aria-label") === "Copy message")).toBe(true);
+    expect(buttons.some((node) => node.getAttribute("aria-label") === "Edit message")).toBe(true);
+    expect(buttons.some((node) => node.getAttribute("aria-label") === "Delete message")).toBe(true);
+
+    expect(container.textContent).toContain("Sent 18");
+    expect(container.textContent).toContain("Generated 24");
+    expect(container.textContent).toContain("Total 42");
+    expect(container.textContent).toContain("6.4 tok/s");
+  });
+});

tauri-app/src/components/MessageBubble.tsx

@@ -14,10 +14,9 @@ export default function MessageBubble({ message, onDelete, onEditAndResend, onEd
   const [editing, setEditing] = useState(false);
   const [editValue, setEditValue] = useState(message.content);
   const [copied, setCopied] = useState(false);
-  const [showReasoning, setShowReasoning] = useState(false);
   const textareaRef = useRef<HTMLTextAreaElement>(null);
   const bubbleRef = useRef<HTMLDivElement>(null);
-  const reasoningRef = useRef<HTMLDivElement>(null);
+  const thinkingRef = useRef<HTMLDivElement>(null);
 
   // Wire up copy-code buttons inside rendered markdown
   useEffect(() => {
@@ -36,8 +35,8 @@ export default function MessageBubble({ message, onDelete, onEditAndResend, onEd
   });
 
   useEffect(() => {
-    if (!reasoningRef.current) return;
-    const buttons = reasoningRef.current.querySelectorAll<HTMLButtonElement>(".copy-code-btn");
+    if (!thinkingRef.current) return;
+    const buttons = thinkingRef.current.querySelectorAll<HTMLButtonElement>(".copy-code-btn");
     buttons.forEach((btn) => {
       btn.onclick = () => {
         const code = btn.getAttribute("data-code") || "";
@@ -90,6 +89,8 @@ export default function MessageBubble({ message, onDelete, onEditAndResend, onEd
     (message.total_tokens ?? 0) > 0
   );
 
+  const showFooter = message.role === "assistant";
+
   return (
     <div className={`msg-wrapper ${message.role}`}>
       {editing ? (
@@ -118,56 +119,101 @@ export default function MessageBubble({ message, onDelete, onEditAndResend, onEd
         </div>
       ) : (
         <>
-          {hasReasoning && (
-            <div className="reasoning-panel">
-              <button
-                className="reasoning-toggle"
-                onClick={() => setShowReasoning((value) => !value)}
-                type="button"
-              >
-                <span>{showReasoning ? "▼" : "▶"}</span>
-                <span>Thinking</span>
-              </button>
-              {showReasoning && (
-                <div
-                  ref={reasoningRef}
-                  className="reasoning-content"
-                  dangerouslySetInnerHTML={{ __html: renderMarkdown(message.reasoning_content || "") }}
-                />
-              )}
-            </div>
-          )}
-
           {message.role === "assistant" ? (
             <div
               ref={bubbleRef}
               className="msg-bubble"
-              dangerouslySetInnerHTML={{ __html: renderMarkdown(message.content) }}
-            />
-          ) : (
-              <div className="msg-bubble">{message.content}</div>
-            )}
+            >
+              {hasReasoning && (
+                <details className="msg-thinking">
+                  <summary className="reasoning-toggle" aria-label="Toggle thinking details">
+                    <svg className="thinking-chevron" viewBox="0 0 16 16" focusable="false" aria-hidden="true">
+                      <path d="M6 4l4 4-4 4" />
+                    </svg>
+                    Thinking
+                  </summary>
+                  <div
+                    className="thinking-content"
+                    ref={thinkingRef}
+                    dangerouslySetInnerHTML={{ __html: renderMarkdown(message.reasoning_content || "") }}
+                  />
+                </details>
+              )}
 
-          {hasUsage && (
-            <div className="message-metrics">
-              <span>Sent {message.prompt_tokens ?? 0}</span>
-              <span>Generated {message.completion_tokens ?? 0}</span>
-              <span>Total {message.total_tokens ?? 0}</span>
-              {message.tokens_per_second ? <span>{message.tokens_per_second.toFixed(1)} tok/s</span> : null}
+              <div
+                className="msg-content"
+                dangerouslySetInnerHTML={{ __html: renderMarkdown(message.content) }}
+              />
+
+              {showFooter && (
+                <div className="msg-footer">
+                  {hasUsage && (
+                    <div className="message-metrics">
+                      <span>Sent {message.prompt_tokens ?? 0}</span>
+                      <span>Generated {message.completion_tokens ?? 0}</span>
+                      <span>Total {message.total_tokens ?? 0}</span>
+                      {message.tokens_per_second ? <span>{message.tokens_per_second.toFixed(1)} tok/s</span> : null}
+                    </div>
+                  )}
+                  {!hasUsage && <div />}
+                  <div className="msg-actions" aria-label="Message actions">
+                    <button type="button" className={`action-btn ${copied ? "copied" : ""}`} onClick={copyMessage} aria-label="Copy message">
+                      <svg viewBox="0 0 20 20" focusable="false" aria-hidden="true">
+                        <rect x="7" y="7" width="10" height="10" rx="2" />
+                        <path d="M13 7V5a2 2 0 0 0-2-2H5a2 2 0 0 0-2 2v6a2 2 0 0 0 2 2h2" />
+                      </svg>
+                    </button>
+                    <button
+                      type="button"
+                      className="action-btn"
+                      onClick={startEdit}
+                      aria-label={onEditAndResend ? "Edit message and resend" : "Edit message"}
+                    >
+                      <svg viewBox="0 0 20 20" focusable="false" aria-hidden="true">
+                        <path d="M13.586 3.414a2 2 0 0 1 2.828 2.828l-9.17 9.17a1 1 0 0 1-.47.264l-3.774.944a.5.5 0 0 1-.608-.608l.944-3.774a1 1 0 0 1 .264-.47l9.986-9.354z" />
+                      </svg>
+                    </button>
+                    <button type="button" className="action-btn danger" onClick={() => onDelete(message.id)} aria-label="Delete message">
+                      <svg viewBox="0 0 20 20" focusable="false" aria-hidden="true">
+                        <path d="M6 5h8l-.7 10.3a1 1 0 0 1-1 .7H7.7a1 1 0 0 1-1-.7L6 5z" />
+                        <path d="M4 5h12M8.5 5V3.5a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 .5.5V5" />
+                      </svg>
+                    </button>
+                  </div>
+                </div>
+              )}
             </div>
+          ) : (
+            <>
+              <div className="msg-bubble user-bubble">
+                <div className="msg-content">{message.content}</div>
+              </div>
+              <div className="msg-actions compact-actions" aria-label="Message actions">
+                <button type="button" className={`action-btn ${copied ? "copied" : ""}`} onClick={copyMessage} aria-label="Copy message">
+                  <svg viewBox="0 0 20 20" focusable="false" aria-hidden="true">
+                    <rect x="7" y="7" width="10" height="10" rx="2" />
+                    <path d="M13 7V5a2 2 0 0 0-2-2H5a2 2 0 0 0-2 2v6a2 2 0 0 0 2 2h2" />
+                  </svg>
+                </button>
+                <button
+                  type="button"
+                  className="action-btn"
+                  onClick={startEdit}
+                  aria-label={onEditAndResend ? "Edit message and resend" : "Edit message"}
+                >
+                  <svg viewBox="0 0 20 20" focusable="false" aria-hidden="true">
+                    <path d="M13.586 3.414a2 2 0 0 1 2.828 2.828l-9.17 9.17a1 1 0 0 1-.47.264l-3.774.944a.5.5 0 0 1-.608-.608l.944-3.774a1 1 0 0 1 .264-.47l9.986-9.354z" />
+                  </svg>
+                </button>
+                <button type="button" className="action-btn danger" onClick={() => onDelete(message.id)} aria-label="Delete message">
+                  <svg viewBox="0 0 20 20" focusable="false" aria-hidden="true">
+                    <path d="M6 5h8l-.7 10.3a1 1 0 0 1-1 .7H7.7a1 1 0 0 1-1-.7L6 5z" />
+                    <path d="M4 5h12M8.5 5V3.5a.5.5 0 0 1 .5-.5h2a.5.5 0 0 1 .5.5V5" />
+                  </svg>
+                </button>
+              </div>
+            </>
           )}
-
-          <div className="msg-actions">
-            <button className={`action-btn ${copied ? "copied" : ""}`} onClick={copyMessage}>
-              {copied ? "✓ Copied" : "⎘ Copy"}
-            </button>
-            <button className="action-btn" onClick={startEdit}>
-              ✎ Edit{onEditAndResend ? " & Resend" : ""}
-            </button>
-            <button className="action-btn danger" onClick={() => onDelete(message.id)}>
-              ✕ Delete
-            </button>
-          </div>
         </>
       )}
     </div>

tauri-app/src/components/ModelSelectorPill.tsx

@@ -0,0 +1,102 @@
+import { useState, useRef, useEffect } from "react";
+import { ChevronDown, Loader2 } from "lucide-react";
+import type { ModelInfo, ServerStatus } from "../lib/types";
+import type { RuntimeModelFile } from "../features/runtime/types";
+import { getDisplayName, isInChatSelector } from "../lib/modelAlias";
+
+interface ModelSelectorPillProps {
+  models: ModelInfo[];
+  localModels: RuntimeModelFile[];
+  selectedModel: string;
+  onModelChange: (id: string) => void;
+  onModelSwap?: (path: string) => void;
+  isSwapping?: boolean;
+  status: ServerStatus;
+}
+
+export default function ModelSelectorPill({
+  models,
+  localModels,
+  selectedModel,
+  onModelChange,
+  onModelSwap,
+  isSwapping,
+  status,
+}: ModelSelectorPillProps) {
+  const [dropdownOpen, setDropdownOpen] = useState(false);
+  const dropdownRef = useRef<HTMLDivElement>(null);
+
+  useEffect(() => {
+    if (!dropdownOpen) return;
+    const handleClickOutside = (e: MouseEvent) => {
+      if (dropdownRef.current && !dropdownRef.current.contains(e.target as Node)) {
+        setDropdownOpen(false);
+      }
+    };
+    document.addEventListener("mousedown", handleClickOutside);
+    return () => document.removeEventListener("mousedown", handleClickOutside);
+  }, [dropdownOpen]);
+
+  const handleSelectModel = (modelId: string) => {
+    setDropdownOpen(false);
+    onModelChange(modelId);
+    if (onModelSwap) {
+      onModelSwap(modelId);
+    }
+  };
+
+  // Unified model list: merge API models + local scanned files
+  const allAvailableModels = [
+    ...models.map((m) => ({ id: m.id, label: getDisplayName(m.id) })),
+    ...localModels.map((m) => ({ id: m.path, label: getDisplayName(m.path) })),
+  ];
+  // Deduplicate by id
+  const seen = new Set<string>();
+  const dedupedModels = allAvailableModels.filter((m) => {
+    if (seen.has(m.id)) return false;
+    seen.add(m.id);
+    return true;
+  });
+
+  // Chat selector: only Cooper + Iron (Steel is settings-only)
+  const chatSelectorModels = dedupedModels.filter((m) => isInChatSelector(m.id));
+
+  const selectedLabel = getDisplayName(selectedModel) || "No model";
+
+  return (
+    <div className="model-selector-pill-wrapper" ref={dropdownRef}>
+      <button
+        type="button"
+        className="model-selector-pill"
+        onClick={() => setDropdownOpen(!dropdownOpen)}
+        disabled={chatSelectorModels.length === 0 || isSwapping}
+        title={chatSelectorModels.length === 0 ? "Start server to load models" : ""}
+        aria-label="Choose active model"
+        aria-expanded={dropdownOpen}
+      >
+        <span className={`model-status-dot status-${status}`} aria-hidden="true" />
+        {isSwapping ? (
+          <Loader2 size={12} strokeWidth={2} className="animate-spin" />
+        ) : (
+          <span className="model-selector-label">{selectedLabel}</span>
+        )}
+        <ChevronDown size={12} strokeWidth={2} className="model-selector-chevron" />
+      </button>
+      {dropdownOpen && (
+        <ul className="model-selector-menu" role="listbox">
+          {chatSelectorModels.map((m) => (
+            <li
+              key={m.id}
+              role="option"
+              aria-selected={m.id === selectedModel}
+              className={`model-selector-item${m.id === selectedModel ? " model-selector-item--selected" : ""}`}
+              onClick={() => handleSelectModel(m.id)}
+            >
+              {m.label}
+            </li>
+          ))}
+        </ul>
+      )}
+    </div>
+  );
+}

tauri-app/src/components/Sidebar.css

@@ -1,19 +1,36 @@
 .sidebar {
-  width: 280px;
-  min-width: 280px;
-  background:
-    radial-gradient(circle at top left, rgba(72, 166, 255, 0.12), transparent 26%),
-    var(--bg2);
-  border-right: 1px solid var(--border);
+  width: 276px;
+  min-width: 276px;
+  background: #1f1f1f;
+  border-right: 1px solid rgba(255, 255, 255, 0.04);
   display: flex;
   flex-direction: column;
   overflow: hidden;
 }
 
+.sidebar--initial .model-section-label,
+.sidebar--initial .footer-meta,
+.sidebar--initial .footer-version {
+  display: none;
+}
+
+.sidebar--initial .model-section {
+  padding-top: 8px;
+}
+
+.sidebar--initial .model-row .icon-btn {
+  display: none;
+}
+
+.sidebar--initial .model-dropdown-trigger {
+  background: transparent;
+  border-color: rgba(255, 255, 255, 0.03);
+}
+
 .sidebar-header-actions {
   display: flex;
   align-items: center;
-  gap: 6px;
+  gap: 4px;
 }
 
 .sidebar-close-btn {
@@ -23,9 +40,9 @@
 /* ── Header ── */
 .sidebar-header {
   display: flex;
-  align-items: center;
-  padding: 14px 14px 12px;
-  border-bottom: 1px solid var(--border);
+  align-items: flex-start;
+  padding: 14px 12px 10px;
+  border-bottom: 0;
   gap: 8px;
   flex-shrink: 0;
 }
@@ -33,32 +50,83 @@
 .sidebar-brand {
   display: flex;
   align-items: center;
-  gap: 9px;
+  gap: 10px;
   flex: 1;
   min-width: 0;
 }
 
-.brand-icon {
-  width: 28px;
-  height: 28px;
-  background: linear-gradient(135deg, #2b88e6, #49c0ff);
+.brand-mark {
+  width: 26px;
+  height: 26px;
   border-radius: 8px;
-  display: flex;
+  border: 1px solid rgba(255, 255, 255, 0.05);
+  background: rgba(255, 255, 255, 0.02);
+  color: var(--text);
+  display: inline-flex;
   align-items: center;
   justify-content: center;
-  font-size: 14px;
+  font-size: 0.82rem;
+  font-weight: 700;
   flex-shrink: 0;
-  box-shadow: 0 2px 8px rgba(91,106,240,0.35);
+}
+
+.brand-copy {
+  display: flex;
+  flex-direction: column;
+  min-width: 0;
 }
 
 .brand-name {
   font-size: 15px;
-  font-weight: 800;
-  letter-spacing: -0.02em;
-  background: linear-gradient(135deg, #f5fbff, #89b5d9);
-  -webkit-background-clip: text;
-  -webkit-text-fill-color: transparent;
-  background-clip: text;
+  font-weight: 600;
+  letter-spacing: -0.01em;
+  color: var(--text);
+}
+
+.brand-subtitle {
+  display: none;
+}
+
+.sidebar-nav-section {
+  padding: 2px 8px 10px;
+  flex-shrink: 0;
+}
+
+.sidebar-section-label {
+  display: none;
+}
+
+.sidebar-nav-list {
+  display: grid;
+  gap: 4px;
+}
+
+.sidebar-nav-item {
+  width: 100%;
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  border: 1px solid transparent;
+  background: transparent;
+  color: var(--text2);
+  border-radius: 10px;
+  padding: 10px 12px;
+  font: inherit;
+  font-size: 13px;
+  cursor: pointer;
+  text-align: left;
+  transition: background 0.15s, border-color 0.15s, color 0.15s;
+}
+
+.sidebar-nav-item:hover {
+  background: rgba(255, 255, 255, 0.03);
+  border-color: rgba(255, 255, 255, 0.05);
+  color: var(--text);
+}
+
+.sidebar-nav-item svg {
+  color: var(--text3);
+  flex-shrink: 0;
 }
 
 .icon-btn {
@@ -66,78 +134,34 @@
   border: none;
   color: var(--text3);
   cursor: pointer;
-  font-size: 15px;
-  width: 36px;
-  height: 36px;
+  width: 32px;
+  height: 32px;
   display: flex;
   align-items: center;
   justify-content: center;
   border-radius: var(--radius-sm);
   transition: background 0.15s, color 0.15s;
   flex-shrink: 0;
-  line-height: 1;
+  padding: 0;
 }
 
 .icon-btn:hover {
-  background: var(--bg4);
+  background: rgba(255, 255, 255, 0.04);
   color: var(--text);
 }
 
-/* ── Settings panel ── */
-.settings-panel {
-  padding: 10px 12px 12px;
-  border-bottom: 1px solid var(--border);
-  display: flex;
-  flex-direction: column;
-  gap: 5px;
-  background: var(--bg3);
-  animation: slide-down 0.15s ease;
-  flex-shrink: 0;
-}
-
-@keyframes slide-down {
-  from { opacity: 0; transform: translateY(-6px); }
-  to   { opacity: 1; transform: translateY(0); }
-}
-
-.settings-label {
-  font-size: 10px;
-  font-weight: 700;
-  color: var(--text3);
-  text-transform: uppercase;
-  letter-spacing: 0.06em;
-  margin-top: 4px;
-}
-
-.settings-input {
-  background: var(--bg4);
-  border: 1px solid var(--border);
-  border-radius: var(--radius-sm);
-  color: var(--text);
-  font-size: 12px;
-  padding: 7px 10px;
-  outline: none;
-  transition: border-color 0.15s, background 0.15s;
-  font-family: inherit;
-  width: 100%;
-  box-sizing: border-box;
-}
-
-.settings-input:focus {
-  border-color: var(--accent);
-  background: var(--bg5);
-}
-
 /* ── Model section ── */
 .model-section {
-  padding: 8px 12px;
-  border-bottom: 1px solid var(--border);
+  margin-top: auto;
+  padding: 10px 12px 12px;
+  border-top: 1px solid rgba(255, 255, 255, 0.04);
+  border-bottom: 0;
   flex-shrink: 0;
 }
 
 .model-section-label {
-  font-size: 10px;
-  font-weight: 700;
+  font-size: 11px;
+  font-weight: 600;
   color: var(--text3);
   text-transform: uppercase;
   letter-spacing: 0.06em;
@@ -147,14 +171,7 @@
 .model-row {
   display: flex;
   align-items: center;
-  gap: 7px;
-}
-
-.model-help {
-  margin-top: 8px;
-  font-size: 11px;
-  color: var(--text3);
-  line-height: 1.5;
+  gap: 8px;
 }
 
 .status-dot {
@@ -186,56 +203,160 @@
 
 .model-select {
   flex: 1;
-  background: var(--bg4);
-  border: 1px solid var(--border);
+  background: transparent;
+  border: 1px solid transparent;
   border-radius: var(--radius-sm);
   color: var(--text);
-  font-size: 12px;
-  padding: 6px 28px 6px 9px;
+  font-size: 13px;
+  padding: 6px 28px 6px 8px;
   outline: none;
   cursor: pointer;
   min-width: 0;
   font-family: inherit;
-  transition: border-color 0.15s;
+  transition: border-color 0.15s, background 0.15s;
   appearance: none;
   background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='10' height='6' viewBox='0 0 10 6'%3E%3Cpath d='M0 0l5 6 5-6z' fill='%236b7280'/%3E%3C/svg%3E");
   background-repeat: no-repeat;
   background-position: right 8px center;
 }
 
+.model-select:hover {
+  background-color: var(--bg3);
+  border-color: var(--border);
+}
+
 .model-select:focus {
   border-color: var(--accent);
+  background-color: var(--bg3);
+}
+
+.model-select:disabled {
+  opacity: 0.5;
+  cursor: not-allowed;
+}
+
+/* ── Custom dropdown (replaces native select) ── */
+.model-dropdown-wrapper {
+  flex: 1;
+  min-width: 0;
+  position: relative;
+}
+
+.model-dropdown-trigger {
+  width: 100%;
+  background: rgba(255, 255, 255, 0.025);
+  border: 1px solid transparent;
+  border-radius: var(--radius-sm);
+  color: var(--text);
+  font-size: 13px;
+  padding: 6px 28px 6px 8px;
+  outline: none;
+  cursor: pointer;
+  min-width: 0;
+  font-family: inherit;
+  transition: border-color 0.15s, background 0.15s;
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 4px;
+}
+
+.model-dropdown-trigger:hover:not(:disabled) {
+  background: rgba(255, 255, 255, 0.05);
+  border-color: var(--border);
+}
+
+.model-dropdown-trigger:focus {
+  border-color: var(--accent);
+  background: rgba(255, 255, 255, 0.05);
+}
+
+.model-dropdown-trigger:disabled {
+  opacity: 0.5;
+  cursor: not-allowed;
+}
+
+.model-dropdown-chevron {
+  flex-shrink: 0;
+  color: var(--text3);
+  transition: transform 0.15s;
+}
+
+.model-dropdown-trigger[aria-expanded="true"] .model-dropdown-chevron {
+  transform: rotate(180deg);
+}
+
+.model-dropdown-menu {
+  position: absolute;
+  top: calc(100% + 4px);
+  left: 0;
+  right: 0;
+  z-index: 100;
+  background: var(--bg2);
+  border: 1px solid var(--border);
+  border-radius: var(--radius-sm);
+  box-shadow: var(--shadow-lg);
+  max-height: 200px;
+  overflow-y: auto;
+  padding: 4px 0;
+  list-style: none;
+}
+
+.model-dropdown-item {
+  padding: 8px 12px;
+  font-size: 13px;
+  color: var(--text2);
+  cursor: pointer;
+  transition: background 0.1s, color 0.1s;
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+}
+
+.model-dropdown-item:hover {
+  background: rgba(255, 255, 255, 0.04);
+  color: var(--text);
+}
+
+.model-dropdown-item--selected {
+  background: rgba(72,166,255,0.07);
+  color: var(--accent);
+  font-weight: 500;
 }
 
 /* ── Sidebar actions ── */
 .sidebar-actions {
-  padding: 10px 12px 6px;
+  padding: 6px 8px 8px;
   flex-shrink: 0;
 }
 
+.sidebar-actions--top {
+  padding-top: 2px;
+}
+
 .new-chat-btn {
   width: 100%;
-  background: linear-gradient(135deg, #2b88e6, #49c0ff);
-  color: #fff;
-  border: none;
-  border-radius: var(--radius-sm);
-  padding: 9px 12px;
+  background: rgba(255, 255, 255, 0.05);
+  color: var(--text);
+  border: 1px solid rgba(255, 255, 255, 0.02);
+  border-radius: 10px;
+  padding: 10px 13px;
   font-size: 13px;
-  font-weight: 700;
+  font-weight: 500;
   cursor: pointer;
   display: flex;
   align-items: center;
-  justify-content: center;
-  gap: 7px;
-  transition: opacity 0.15s, transform 0.1s, box-shadow 0.15s;
+  justify-content: flex-start;
+  gap: 8px;
+  transition: background 0.15s, color 0.15s, border-color 0.15s, transform 0.15s;
   font-family: inherit;
-  letter-spacing: 0.01em;
-  box-shadow: 0 2px 10px rgba(72,166,255,0.25);
+  height: 40px;
 }
 
 .new-chat-btn:hover {
-  opacity: 0.9;
-  box-shadow: 0 4px 16px rgba(72,166,255,0.4);
+  background: rgba(255, 255, 255, 0.08);
+  color: var(--text);
+  border-color: rgba(255, 255, 255, 0.05);
 }
 
 .new-chat-btn:active {
@@ -245,18 +366,18 @@
 /* ── Chat list ── */
 .chat-list-section-label {
   font-size: 10px;
-  font-weight: 700;
+  font-weight: 600;
   color: var(--text3);
   text-transform: uppercase;
-  letter-spacing: 0.06em;
-  padding: 8px 12px 4px;
+  letter-spacing: 0.12em;
+  padding: 12px 12px 6px;
   flex-shrink: 0;
 }
 
 .chat-list {
   flex: 1;
   overflow-y: auto;
-  padding: 4px 8px 8px;
+  padding: 4px 8px 10px;
   display: flex;
   flex-direction: column;
   gap: 1px;
@@ -266,7 +387,7 @@
   text-align: center;
   color: var(--text3);
   font-size: 12px;
-  padding: 24px 0;
+  padding: 20px 8px 24px;
   line-height: 1.6;
 }
 
@@ -279,7 +400,7 @@
   color: var(--text2);
   transition: background 0.12s, color 0.12s;
   position: relative;
-  border-left: 2px solid transparent;
+  border-top: 1px solid rgba(255, 255, 255, 0.02);
 }
 
 .chat-item-button {
@@ -287,40 +408,29 @@
   min-width: 0;
   display: flex;
   align-items: center;
-  gap: 8px;
-  padding: 8px 10px;
+  padding: 10px 10px;
   background: transparent;
   border: 0;
   color: inherit;
   cursor: pointer;
   font: inherit;
   text-align: left;
+  border-radius: var(--radius-sm);
 }
 
 .chat-item:hover {
-  background: var(--bg3);
+  background: rgba(255, 255, 255, 0.04);
   color: var(--text);
 }
 
 .chat-item:focus-within {
-  background: var(--bg3);
+  background: rgba(255, 255, 255, 0.04);
   color: var(--text);
 }
 
 .chat-item.active {
-  background: rgba(91,106,240,0.1);
+  background: rgba(255, 255, 255, 0.06);
   color: var(--text);
-  border-left-color: var(--accent);
-}
-
-.chat-item-icon {
-  font-size: 13px;
-  flex-shrink: 0;
-  opacity: 0.5;
-}
-
-.chat-item.active .chat-item-icon {
-  opacity: 0.8;
 }
 
 .chat-item-title {
@@ -329,6 +439,7 @@
   text-overflow: ellipsis;
   white-space: nowrap;
   font-size: 13px;
+  line-height: 1.4;
 }
 
 .chat-item-del {
@@ -337,7 +448,6 @@
   border: none;
   color: var(--text3);
   cursor: pointer;
-  font-size: 12px;
   width: 28px;
   height: 28px;
   display: flex;
@@ -346,8 +456,8 @@
   border-radius: 4px;
   transition: opacity 0.15s, background 0.15s, color 0.15s;
   flex-shrink: 0;
-  line-height: 1;
-  margin-right: 6px;
+  padding: 0;
+  margin-right: 4px;
 }
 
 .chat-item:hover .chat-item-del {
@@ -366,20 +476,35 @@
 
 /* ── Footer ── */
 .sidebar-footer {
-  padding: 8px 14px;
-  border-top: 1px solid var(--border);
+  padding: 8px 12px 14px;
+  border-top: 0;
   display: flex;
   align-items: center;
   justify-content: space-between;
   flex-shrink: 0;
 }
 
+.sidebar--initial .sidebar-footer {
+  padding-top: 0;
+}
+
+.sidebar-footer-copy {
+  display: flex;
+  flex-direction: column;
+  gap: 2px;
+}
+
 .footer-version {
   font-size: 10px;
   color: var(--text3);
   letter-spacing: 0.02em;
 }
 
+.footer-meta {
+  font-size: 11px;
+  color: var(--text2);
+}
+
 @media (max-width: 960px) {
   .sidebar {
     position: fixed;

tauri-app/src/components/Sidebar.tsx

@@ -1,73 +1,51 @@
-import { useState } from "react";
-import type { Chat, ModelInfo, ServerStatus } from "../lib/types";
-import RuntimeWorkspace from "../features/runtime/components/RuntimeWorkspace";
-import type { RuntimeConfig, RuntimeDraftFieldMeta, RuntimeDraftFieldState, RuntimeSnapshot } from "../features/runtime/types";
+import type { Chat } from "../lib/types";
+import { Plus, Settings, X, Trash2, Bot, Search, Library, PanelLeft } from "lucide-react";
 import "./Sidebar.css";
 
 interface Props {
   chats: Chat[];
   currentChatId: number | null;
+  initialScreen?: boolean;
   onSelectChat: (id: number) => void;
   onNewChat: () => void;
   onDeleteChat: (id: number) => void;
-  models: ModelInfo[];
-  selectedModel: string;
-  onModelChange: (v: string) => void;
-  status: ServerStatus;
-  onRefreshModels: () => void;
-  runtime: {
-    snapshot: RuntimeSnapshot;
-    savedBaseline: RuntimeConfig;
-    draftConfig: RuntimeConfig;
-    draftDirty: boolean;
-    draftOpen: boolean;
-    draftFields: RuntimeDraftFieldState;
-    loading: boolean;
-    busy: boolean;
-    activeLifecycleCommand: RuntimeSnapshot["active_command"];
-    lastLifecycleResult: RuntimeSnapshot["last_command_result"];
-    lifecycleBusy: boolean;
-    applyImpact: "immediate" | "restart_required" | "save_before_apply";
-    openDraft: () => void;
-    closeDraft: () => void;
-    resetDraft: () => void;
-    updateDraft: (patch: Partial<RuntimeConfig>) => void;
-    updateDraftFieldMeta: (field: keyof RuntimeDraftFieldState & string, patch: Partial<RuntimeDraftFieldMeta>) => void;
-    saveConfig: (patch: Partial<RuntimeConfig>) => Promise<RuntimeSnapshot>;
-    startLocal: () => Promise<unknown>;
-    stopLocal: () => Promise<unknown>;
-  };
+  onOpenSettings: () => void;
   onCloseMobile: () => void;
 }
 
 export default function Sidebar({
-  chats, currentChatId, onSelectChat, onNewChat, onDeleteChat,
-  models, selectedModel, onModelChange, status, onRefreshModels,
-  runtime,
+  chats, currentChatId, initialScreen = false, onSelectChat, onNewChat, onDeleteChat,
+  onOpenSettings,
   onCloseMobile,
 }: Props) {
-  const modelCountLabel = models.length === 1 ? "1 model available" : `${models.length} models available`;
+  const workspaceNavItems = [
+    { label: "Agents", icon: Bot },
+    { label: "Search", icon: Search },
+    { label: "Library", icon: Library },
+  ];
 
   return (
     <>
-      <aside className="sidebar">
+      <aside className={`sidebar${initialScreen ? " sidebar--initial" : ""}`}>
         {/* ── Header ── */}
         <div className="sidebar-header">
           <div className="sidebar-brand">
-            <div className="brand-icon">⚡</div>
-            <span className="brand-name">ForgeLM</span>
+            <div className="brand-mark" aria-hidden="true">
+              <PanelLeft size={15} strokeWidth={1.8} />
+            </div>
+            <div className="brand-copy">
+              <span className="brand-name">ForgeLM</span>
+            </div>
           </div>
           <div className="sidebar-header-actions">
             <button
               className="icon-btn"
-              onClick={() => {
-                runtime.openDraft();
-              }}
-              title="Runtime settings"
-              aria-label="Open runtime settings"
+              onClick={onOpenSettings}
+              title="Settings"
+              aria-label="Open settings"
               type="button"
             >
-              ⚙
+              <Settings size={16} strokeWidth={1.5} />
             </button>
             <button
               className="icon-btn sidebar-close-btn"
@@ -76,65 +54,35 @@ export default function Sidebar({
               aria-label="Close navigation"
               type="button"
             >
-              ✕
+              <X size={16} strokeWidth={1.5} />
             </button>
           </div>
         </div>
 
-        {/* ── Unified Runtime Workspace ── */}
-        <RuntimeWorkspace runtime={runtime} />
-
-        {/* ── Active model selector ── */}
-        <div className="model-section">
-          <div className="model-section-label">Active model</div>
-          <div className="model-row">
-            <div className={`status-dot status-${status}`} title={status} aria-hidden="true" />
-            <select
-              className="model-select"
-              value={selectedModel}
-              onChange={(e) => onModelChange(e.target.value)}
-              disabled={models.length === 0}
-              title={models.length === 0 ? "Start the server to load models" : ""}
-              aria-label="Choose active model"
-            >
-              {models.length === 0 ? (
-                <option value="">Start server to load models</option>
-              ) : (
-                models.map((m) => (
-                  <option key={m.id} value={m.id}>{m.id}</option>
-                ))
-              )}
-            </select>
-            <button
-              className="icon-btn"
-              onClick={onRefreshModels}
-              title="Refresh models"
-              aria-label="Refresh models"
-              type="button"
-            >
-              ↻
-            </button>
-          </div>
-          <div className="model-help">
-            {models.length === 0 ? "Start the runtime to load available models." : modelCountLabel}
-          </div>
-        </div>
-
-        {/* ── New chat ── */}
-        <div className="sidebar-actions">
+        <div className="sidebar-actions sidebar-actions--top">
           <button className="new-chat-btn" onClick={onNewChat} type="button">
-            <span>＋</span> New Chat
+            <Plus size={16} strokeWidth={1.5} />
+            <span>New task</span>
           </button>
         </div>
 
+        <div className="sidebar-nav-section">
+          <div className="sidebar-nav-list" aria-label="Workspace navigation">
+            {workspaceNavItems.map(({ label, icon: Icon }) => (
+              <button key={label} type="button" className="sidebar-nav-item">
+                <Icon size={15} strokeWidth={1.8} />
+                <span>{label}</span>
+              </button>
+            ))}
+          </div>
+        </div>
+
         {/* ── Chat list ── */}
-        {chats.length > 0 && (
-          <div className="chat-list-section-label">Conversations</div>
-        )}
+        {chats.length > 0 && <div className="chat-list-section-label">Recent work</div>}
         <nav className="chat-list" aria-label="Conversation history">
           {chats.length === 0 && (
             <div className="chat-list-empty">
-              No conversations yet.<br />Start a new chat above.
+              No work yet.<br />Start a new task above.
             </div>
           )}
           {chats.map((chat) => (
@@ -148,7 +96,6 @@ export default function Sidebar({
                 onClick={() => onSelectChat(chat.id)}
                 aria-current={chat.id === currentChatId ? "page" : undefined}
               >
-                <span className="chat-item-icon">💬</span>
                 <span className="chat-item-title">{chat.title}</span>
               </button>
               <button
@@ -158,7 +105,7 @@ export default function Sidebar({
                 title="Delete conversation"
                 aria-label={`Delete conversation ${chat.title}`}
               >
-                ✕
+                <Trash2 size={14} strokeWidth={1.5} />
               </button>
             </div>
           ))}
@@ -166,7 +113,10 @@ export default function Sidebar({
 
         {/* ── Footer ── */}
         <div className="sidebar-footer">
-          <span className="footer-version">ForgeLM v0.1.0</span>
+          <div className="sidebar-footer-copy">
+            <span className="footer-version">ForgeLM v0.1.0</span>
+            <span className="footer-meta">Local-first workspace</span>
+          </div>
         </div>
 
       </aside>

tauri-app/src/features/runtime/api.ts

@@ -1,6 +1,7 @@
 import { invoke } from "@tauri-apps/api/core";
 import { createDefaultRuntimeSnapshot, createRuntimeLifecycleResult } from "./runtime-utils";
 import type {
+  HuggingFaceDownloadProgress,
   ModelLibraryState,
   HuggingFaceModelResult,
   RuntimeConfig,
@@ -11,6 +12,7 @@ import type {
 } from "./types";
 
 const STORAGE_KEY = "forgelm_runtime_snapshot_v1";
+export const HUGGING_FACE_DOWNLOAD_EVENT = "hugging-face-download-progress";
 
 function canUseLocalStorage(): boolean {
   return typeof window !== "undefined" && typeof window.localStorage !== "undefined";
@@ -146,6 +148,62 @@ export async function searchHuggingFaceModels(query: string, token = ""): Promis
   });
 }
 
+export async function startHuggingFaceDownload(repo: string, fileName: string, token = ""): Promise<string> {
+  try {
+    return await invoke<string>("start_hugging_face_download_command", {
+      repo,
+      fileName,
+      token: token.trim() || null,
+    });
+  } catch (err) {
+    console.error("[api] startHuggingFaceDownload failed:", err);
+    throw err;
+  }
+}
+
+export async function cancelHuggingFaceDownload(downloadId: string): Promise<void> {
+  try {
+    return await invoke<void>("cancel_hugging_face_download_command", {
+      downloadId,
+    });
+  } catch (err) {
+    console.error("[api] cancelHuggingFaceDownload failed:", err);
+    throw err;
+  }
+}
+
 export async function getModelLibraryState(): Promise<ModelLibraryState> {
   return invoke<ModelLibraryState>("get_model_library_state_command");
 }
+
+/// Hot-swap: stop runtime → update model_path → restart.
+export async function swapModel(newModelPath: string): Promise<RuntimeSnapshot> {
+  try {
+    return await invoke<RuntimeSnapshot>("swap_model_command", { newModelPath });
+  } catch (err) {
+    console.error("[api] swapModel failed:", err);
+    throw err;
+  }
+}
+
+/// Delete a local .gguf model file from disk.
+export async function deleteModelFile(filePath: string): Promise<void> {
+  try {
+    return await invoke<void>("delete_model_file", { filePath });
+  } catch (err) {
+    console.error("[api] deleteModelFile failed:", err);
+    throw err;
+  }
+}
+
+/// Open the folder containing a model in the native file explorer.
+export async function openModelFolder(filePath: string): Promise<void> {
+  try {
+    return await invoke<void>("open_model_folder_command", { filePath });
+  } catch (err) {
+    console.error("[api] openModelFolder failed:", err);
+    throw err;
+  }
+}
+
+export type { HuggingFaceDownloadProgress };

tauri-app/src/features/runtime/components/RuntimeSettingsDrawer.tsx

@@ -16,6 +16,7 @@ import {
   resolveDraftModelSelection,
   toModelFile,
 } from "../runtime-settings-form";
+import { CTX_PRESETS, DEFAULT_CTX_SIZE, MIN_CTX_SIZE } from "../runtime-defaults";
 import {
   createRuntimeDraftFieldState,
   type ModelLibraryState,
@@ -31,7 +32,6 @@ import {
 } from "../types";
 import "./RuntimeSettingsDrawer.css";
 
-const CTX_PRESETS = [2048, 4096, 8192, 16384, 32768, 65536];
 const REMOTE_PRESET_TEMPLATES = [
   { id: "openai", label: "OpenAI", provider: "open_ai", baseUrl: "https://api.openai.com/v1" },
   { id: "openrouter", label: "OpenRouter", provider: "open_router", baseUrl: "https://openrouter.ai/api/v1" },
@@ -799,10 +799,10 @@ export default function RuntimeSettingsDrawer({
                   <input
                     id="runtime-context"
                     className={`runtime-field-input${isFieldDirty("ctx_size") ? " is-draft" : ""}`}
-                    min={512}
+                    min={MIN_CTX_SIZE}
                     step={512}
                     value={draftConfig.ctx_size}
-                    onChange={(event) => updateDraft({ ctx_size: Math.max(512, parseInt(event.target.value, 10) || 4096) })}
+                    onChange={(event) => updateDraft({ ctx_size: Math.max(MIN_CTX_SIZE, parseInt(event.target.value, 10) || DEFAULT_CTX_SIZE) })}
                     type="number"
                   />
                 </div>

tauri-app/src/features/runtime/components/RuntimeStatusCard.css

@@ -28,6 +28,11 @@
   gap: 10px;
 }
 
+.runtime-card-header-left {
+  min-width: 0;
+  flex: 1;
+}
+
 .runtime-card-kicker {
   font-size: 10px;
   font-weight: 700;
@@ -42,6 +47,25 @@
   font-weight: 700;
   letter-spacing: -0.02em;
   color: var(--text);
+  display: flex;
+  align-items: center;
+  gap: 7px;
+}
+
+/* Pulsing green dot when runtime is running */
+.runtime-status-dot {
+  display: inline-block;
+  width: 8px;
+  height: 8px;
+  border-radius: 50%;
+  background: #30d158;
+  flex-shrink: 0;
+  animation: runtime-pulse 2s ease-in-out infinite;
+}
+
+@keyframes runtime-pulse {
+  0%, 100% { box-shadow: 0 0 0 0 rgba(48, 209, 88, 0.5); }
+  50%       { box-shadow: 0 0 0 5px rgba(48, 209, 88, 0); }
 }
 
 .runtime-card-mode {
@@ -92,6 +116,13 @@
   padding: 7px 10px;
 }
 
+.runtime-card-meta-model {
+  max-width: 160px;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
 .runtime-card-note {
   display: block;
   margin-top: 10px;
@@ -177,12 +208,63 @@
   border-color: var(--accent);
 }
 
-.runtime-card-btn-primary {
+/* Start — green */
+.runtime-card-btn-start {
+  color: #fff;
+  border-color: rgba(48, 209, 88, 0.5);
+  background: linear-gradient(135deg, #1ea84d, #30d158);
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap: 6px;
+}
+
+.runtime-card-btn-start:hover:not(:disabled) {
+  background: linear-gradient(135deg, #25c05a, #3fe068);
+}
+
+/* Stop — red */
+.runtime-card-btn-stop {
+  color: #fff;
+  border-color: rgba(240, 64, 64, 0.5);
+  background: linear-gradient(135deg, #c0302e, #f04040);
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap: 6px;
+}
+
+.runtime-card-btn-stop:hover:not(:disabled) {
+  background: linear-gradient(135deg, #d03838, #f55555);
+}
+
+/* Setup — blue (same as primary) */
+.runtime-card-btn-setup {
   color: #fff;
   border-color: rgba(72, 166, 255, 0.6);
   background: linear-gradient(135deg, #2b88e6, #46c2ff);
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap: 6px;
 }
 
-.runtime-card-btn-primary:hover:not(:disabled) {
+.runtime-card-btn-setup:hover:not(:disabled) {
   background: linear-gradient(135deg, #3193f5, #55cbff);
 }
+
+/* Spinner for in-progress lifecycle */
+.runtime-btn-spinner {
+  display: inline-block;
+  width: 11px;
+  height: 11px;
+  border: 2px solid rgba(255, 255, 255, 0.35);
+  border-top-color: #fff;
+  border-radius: 50%;
+  animation: runtime-spin 0.7s linear infinite;
+  flex-shrink: 0;
+}
+
+@keyframes runtime-spin {
+  to { transform: rotate(360deg); }
+}

tauri-app/src/features/runtime/components/RuntimeStatusCard.tsx

@@ -62,26 +62,35 @@ export default function RuntimeStatusCard({
         ? "runtime-card-result-warning"
         : "runtime-card-result-success";
 
+  const isRunning = isLocal && snapshot.process_running;
+  const isStopping = isLocal && snapshot.status === "stopping";
+  const primaryVariant = needsSetup ? "setup" : isRunning ? "stop" : "start";
+
   return (
     <section className={`runtime-card runtime-card-${tone}`} aria-label="Runtime manager summary">
       <div className="runtime-card-header">
-        <div>
+        <div className="runtime-card-header-left">
           <div className="runtime-card-kicker">Runtime Manager</div>
-          <div className="runtime-card-title">{loading ? "Loading runtime" : getRuntimeStateLabel(snapshot)}</div>
+          <div className="runtime-card-title">
+            {isRunning && !isStopping && <span className="runtime-status-dot" aria-hidden="true" />}
+            {loading ? "Loading runtime" : getRuntimeStateLabel(snapshot)}
+          </div>
         </div>
         <span className={`runtime-card-mode runtime-card-mode-${snapshot.config.connection_mode}`}>
           {snapshot.config.connection_mode === "local" ? "Local" : "Remote"}
         </span>
       </div>
 
-      <p className="runtime-card-copy">
+      <p className="runtime-card-copy" title={snapshot.config.connection_mode === "local" ? snapshot.resolved_url ?? "" : snapshot.config.server_url ?? ""}>
         {snapshot.config.connection_mode === "local"
           ? `Managed endpoint ${snapshot.resolved_url}`
           : snapshot.config.server_url || "Configure a remote API endpoint"}
       </p>
 
       <div className="runtime-card-meta" aria-label="Runtime details">
-        <span>{isLocal ? summarizePath(snapshot.config.model_path) : "External provider"}</span>
+        <span className="runtime-card-meta-model" title={snapshot.config.model_path ?? ""}>
+          {isLocal ? summarizePath(snapshot.config.model_path) : "External provider"}
+        </span>
         <span>{snapshot.config.start_policy === "on_app_launch" ? "Auto-start" : "Manual start"}</span>
       </div>
 
@@ -104,7 +113,13 @@ export default function RuntimeStatusCard({
           Manage
         </button>
         {snapshot.config.connection_mode === "local" && (
-          <button className="runtime-card-btn runtime-card-btn-primary" onClick={primaryAction} type="button" disabled={primaryDisabled}>
+          <button
+            className={`runtime-card-btn runtime-card-btn-${primaryVariant}`}
+            onClick={primaryAction}
+            type="button"
+            disabled={primaryDisabled}
+          >
+            {lifecycleBusy ? <span className="runtime-btn-spinner" aria-hidden="true" /> : null}
             {primaryLabel}
           </button>
         )}

tauri-app/src/features/runtime/runtime-defaults.ts

@@ -0,0 +1,3 @@
+export const MIN_CTX_SIZE = 512;
+export const DEFAULT_CTX_SIZE = 4096;
+export const CTX_PRESETS = [2048, 4096, 8192, 16384, 32768, 65536, 131072];

tauri-app/src/features/runtime/runtime-settings-form.test.ts

@@ -9,6 +9,7 @@ import {
   toModelFile,
   type RuntimeSettingsFormState,
 } from "./runtime-settings-form";
+import { DEFAULT_CTX_SIZE } from "./runtime-defaults";
 import { createDefaultRuntimeSnapshot } from "./runtime-utils";
 import {
   createRuntimeDraftState,
@@ -185,6 +186,17 @@ describe("runtime settings form helpers", () => {
     });
   });
 
+  it("falls back to the shared desktop default context size when the draft value is empty", () => {
+    const snapshot = createDefaultRuntimeSnapshot();
+    const patch = createRuntimeConfigPatch(
+      snapshot.config,
+      createFormState({ ctxSize: 0 }),
+      32,
+    );
+
+    expect(patch.ctx_size).toBe(DEFAULT_CTX_SIZE);
+  });
+
   it("keeps saved local values untouched when editing remote mode", () => {
     const snapshot = createDefaultRuntimeSnapshot();
     const patch = createRuntimeConfigPatch(

tauri-app/src/features/runtime/runtime-settings-form.ts

@@ -4,6 +4,7 @@ import type {
   RuntimeModelFile,
   StartPolicy,
 } from "./types";
+import { DEFAULT_CTX_SIZE, MIN_CTX_SIZE } from "./runtime-defaults";
 
 function normalizeUrl(value: string): string {
   const trimmed = value.trim();
@@ -126,7 +127,7 @@ export function createRuntimeConfigPatch(
       form.mode === "local"
         ? Math.max(0, Math.min(form.gpuLayers, maxLayers))
         : snapshotConfig.gpu_layers,
-    ctx_size: form.mode === "local" ? Math.max(512, form.ctxSize || 4096) : snapshotConfig.ctx_size,
+    ctx_size: form.mode === "local" ? Math.max(MIN_CTX_SIZE, form.ctxSize || DEFAULT_CTX_SIZE) : snapshotConfig.ctx_size,
     selected_model_id: form.mode === "local" ? form.selectedModelId ?? null : snapshotConfig.selected_model_id ?? null,
     extra_model_paths:
       form.mode === "local"

tauri-app/src/features/runtime/runtime-utils.test.ts

@@ -1,5 +1,13 @@
 import { describe, expect, it } from "vitest";
-import { createRuntimeLifecycleResult, getRuntimeConnection, getRuntimeStartGuard, getRuntimeStateLabel } from "./runtime-utils";
+import { DEFAULT_CTX_SIZE } from "./runtime-defaults";
+import {
+  createDefaultRuntimeSnapshot,
+  createRuntimeLifecycleResult,
+  getRuntimeConnection,
+  getRuntimeStartGuard,
+  getRuntimeStateLabel,
+  getRuntimeStatusTone,
+} from "./runtime-utils";
 import type { RuntimeConfig, RuntimeSnapshot } from "./types";
 
 type SnapshotOverrides = Omit<Partial<RuntimeSnapshot>, "config"> & {
@@ -42,6 +50,10 @@ function createSnapshot(overrides: SnapshotOverrides = {}): RuntimeSnapshot {
 }
 
 describe("getRuntimeConnection", () => {
+  it("creates local snapshots with the shared desktop default context window", () => {
+    expect(createDefaultRuntimeSnapshot().config.ctx_size).toBe(DEFAULT_CTX_SIZE);
+  });
+
   it("prefers the resolved local runtime url in local mode", () => {
     const connection = getRuntimeConnection(createSnapshot({
       resolved_url: "http://127.0.0.1:9000",
@@ -90,6 +102,10 @@ describe("getRuntimeStateLabel", () => {
     expect(getRuntimeStateLabel(createSnapshot({ status: "ready", ready: true, process_running: true }))).toBe("Ready");
   });
 
+  it("reports local processes that lost readiness as not ready", () => {
+    expect(getRuntimeStateLabel(createSnapshot({ status: "ready", ready: false, process_running: true }))).toBe("Not ready");
+  });
+
   it("reports local background start progress", () => {
     expect(getRuntimeStateLabel(createSnapshot({ status: "starting" }))).toBe("Starting local API");
   });
@@ -113,6 +129,12 @@ describe("getRuntimeStateLabel", () => {
   });
 });
 
+describe("getRuntimeStatusTone", () => {
+  it("marks local not-ready snapshots as danger", () => {
+    expect(getRuntimeStatusTone(createSnapshot({ status: "ready", ready: false, process_running: true }))).toBe("danger");
+  });
+});
+
 describe("getRuntimeStartGuard", () => {
   it("requires a saved model before starting the local runtime", () => {
     const guard = getRuntimeStartGuard(createSnapshot({

tauri-app/src/features/runtime/runtime-utils.ts

@@ -13,6 +13,7 @@ import type {
   RuntimeWorkspaceSummary,
 } from "./types";
 import { scanRuntimeModels } from "./api";
+import { DEFAULT_CTX_SIZE } from "./runtime-defaults";
 
 const DEFAULT_LOCAL_URL = "http://127.0.0.1:8080";
 
@@ -179,6 +180,10 @@ export function getRuntimeStateLabel(snapshot: RuntimeSnapshot): string {
     return "Remote API";
   }
 
+  if (snapshot.status === "ready" && !snapshot.ready) {
+    return snapshot.process_running ? "Not ready" : "Local API offline";
+  }
+
   switch (snapshot.status) {
     case "ready":
       return "Ready";
@@ -200,6 +205,10 @@ export function getRuntimeStatusTone(snapshot: RuntimeSnapshot): "neutral" | "su
     return "neutral";
   }
 
+  if (snapshot.status === "ready" && !snapshot.ready) {
+    return snapshot.process_running ? "danger" : "neutral";
+  }
+
   switch (snapshot.status) {
     case "ready":
       return "success";
@@ -299,7 +308,7 @@ export function createDefaultRuntimeSnapshot(): RuntimeSnapshot {
       models_dir: null,
       addr: "127.0.0.1:8080",
       gpu_layers: 0,
-      ctx_size: 4096,
+      ctx_size: DEFAULT_CTX_SIZE,
       extra_model_paths: [],
       selected_model_id: null,
       remote_presets: [],

tauri-app/src/features/runtime/types.ts

@@ -117,6 +117,7 @@ export interface RuntimeConnection {
 export interface RuntimeModelFile {
   name: string;
   path: string;
+  size?: number | null;
 }
 
 export interface RuntimePickDirResult {
@@ -142,6 +143,22 @@ export interface HuggingFaceModelResult {
   verified_files: HuggingFaceVerifiedFile[];
 }
 
+export type HuggingFaceDownloadPhase = "started" | "progress" | "completed" | "cancelled" | "error";
+
+export interface HuggingFaceDownloadProgress {
+  downloadId: string;
+  repo: string;
+  fileName: string;
+  phase: HuggingFaceDownloadPhase;
+  progressPercentage: number;
+  downloadedMB: number;
+  totalMB: number | null;
+  downloadSpeed: number;
+  destinationPath: string | null;
+  message: string | null;
+  error: string | null;
+}
+
 export interface ModelLibraryEntry {
   id: string;
   label: string;

tauri-app/src/features/runtime/useModelDownload.test.tsx

@@ -0,0 +1,156 @@
+/* @vitest-environment jsdom */
+
+import { createRoot, type Root } from "react-dom/client";
+import { act } from "react-dom/test-utils";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+import { HUGGING_FACE_DOWNLOAD_EVENT, type HuggingFaceDownloadProgress } from "./api";
+import { useModelDownload } from "./useModelDownload";
+
+const invokeMock = vi.fn();
+const listenMock = vi.fn();
+
+vi.mock("@tauri-apps/api/core", () => ({
+  invoke: (...args: unknown[]) => invokeMock(...args),
+}));
+
+vi.mock("@tauri-apps/api/event", () => ({
+  listen: (...args: unknown[]) => listenMock(...args),
+}));
+
+function Harness({ token = "hf_token" }: { token?: string }) {
+  const state = useModelDownload(token);
+
+  return (
+    <div>
+      <output id="is-downloading">{String(state.isDownloading)}</output>
+      <output id="progress">{state.progress.toFixed(1)}</output>
+      <output id="speed">{state.speed.toFixed(1)}</output>
+      <output id="downloaded">{state.downloaded.toFixed(1)}</output>
+      <output id="total">{state.total.toFixed(1)}</output>
+      <output id="active-file">{state.activeFile ?? ""}</output>
+      <output id="error">{state.error ?? ""}</output>
+      <button id="start" onClick={() => void state.startDownload("microsoft/phi-4", "phi-4-q4.gguf")}>
+        Start
+      </button>
+      <button id="cancel" onClick={() => void state.cancelDownload()}>
+        Cancel
+      </button>
+    </div>
+  );
+}
+
+describe("useModelDownload", () => {
+  let container: HTMLDivElement;
+  let root: Root;
+  let progressListener: ((event: { payload: HuggingFaceDownloadProgress }) => void) | null = null;
+
+  beforeEach(() => {
+    // @ts-expect-error React test environment flag
+    globalThis.IS_REACT_ACT_ENVIRONMENT = true;
+    container = document.createElement("div");
+    document.body.appendChild(container);
+    root = createRoot(container);
+    invokeMock.mockReset();
+    listenMock.mockReset();
+    progressListener = null;
+    listenMock.mockImplementation(async (eventName, callback) => {
+      if (eventName === HUGGING_FACE_DOWNLOAD_EVENT) {
+        progressListener = callback as (event: { payload: HuggingFaceDownloadProgress }) => void;
+      }
+      return () => undefined;
+    });
+  });
+
+  afterEach(() => {
+    act(() => root.unmount());
+    container.remove();
+    // @ts-expect-error React test environment flag
+    globalThis.IS_REACT_ACT_ENVIRONMENT = false;
+  });
+
+  it("starts a Hugging Face download with the active token and tracks progress events", async () => {
+    invokeMock.mockResolvedValue("download-1");
+
+    await act(async () => {
+      root.render(<Harness token="hf_secret" />);
+    });
+
+    await act(async () => {
+      (container.querySelector("#start") as HTMLButtonElement).click();
+      await Promise.resolve();
+    });
+
+    expect(invokeMock).toHaveBeenCalledWith("start_hugging_face_download_command", {
+      repo: "microsoft/phi-4",
+      fileName: "phi-4-q4.gguf",
+      token: "hf_secret",
+    });
+    expect(container.querySelector("#is-downloading")?.textContent).toBe("true");
+    expect(container.querySelector("#active-file")?.textContent).toBe("microsoft/phi-4:phi-4-q4.gguf");
+
+    await act(async () => {
+      progressListener?.({
+        payload: {
+          downloadId: "download-1",
+          repo: "microsoft/phi-4",
+          fileName: "phi-4-q4.gguf",
+          phase: "progress",
+          progressPercentage: 37.5,
+          downloadedMB: 768,
+          totalMB: 2048,
+          downloadSpeed: 12.4,
+          destinationPath: null,
+          message: null,
+          error: null,
+        },
+      });
+    });
+
+    expect(container.querySelector("#progress")?.textContent).toBe("37.5");
+    expect(container.querySelector("#speed")?.textContent).toBe("12.4");
+    expect(container.querySelector("#downloaded")?.textContent).toBe("768.0");
+    expect(container.querySelector("#total")?.textContent).toBe("2048.0");
+  });
+
+  it("ignores progress for stale downloads and cancels the active one", async () => {
+    invokeMock.mockResolvedValueOnce("download-1").mockResolvedValueOnce(undefined);
+
+    await act(async () => {
+      root.render(<Harness />);
+    });
+
+    await act(async () => {
+      (container.querySelector("#start") as HTMLButtonElement).click();
+      await Promise.resolve();
+    });
+
+    await act(async () => {
+      progressListener?.({
+        payload: {
+          downloadId: "download-stale",
+          repo: "other/repo",
+          fileName: "other.gguf",
+          phase: "progress",
+          progressPercentage: 90,
+          downloadedMB: 90,
+          totalMB: 100,
+          downloadSpeed: 4,
+          destinationPath: null,
+          message: null,
+          error: null,
+        },
+      });
+    });
+
+    expect(container.querySelector("#progress")?.textContent).toBe("0.0");
+
+    await act(async () => {
+      (container.querySelector("#cancel") as HTMLButtonElement).click();
+      await Promise.resolve();
+    });
+
+    expect(invokeMock).toHaveBeenLastCalledWith("cancel_hugging_face_download_command", {
+      downloadId: "download-1",
+    });
+  });
+});

tauri-app/src/features/runtime/useModelDownload.ts

@@ -0,0 +1,164 @@
+import { useCallback, useEffect, useRef, useState } from "react";
+import { listen } from "@tauri-apps/api/event";
+import {
+  HUGGING_FACE_DOWNLOAD_EVENT,
+  cancelHuggingFaceDownload,
+  startHuggingFaceDownload,
+  type HuggingFaceDownloadProgress,
+} from "./api";
+
+interface UseModelDownloadState {
+  isDownloading: boolean;
+  progress: number;
+  speed: number;
+  downloaded: number;
+  total: number;
+  activeFile: string | null;
+  error: string | null;
+  didComplete: boolean;
+  downloadedFilePath: string | null;
+  resetDidComplete: () => void;
+  startDownload: (repo: string, fileName: string) => Promise<void>;
+  cancelDownload: () => Promise<void>;
+}
+
+interface ModelDownloadSnapshot {
+  isDownloading: boolean;
+  progress: number;
+  speed: number;
+  downloaded: number;
+  total: number;
+  activeFile: string | null;
+  error: string | null;
+}
+
+const INITIAL_STATE: ModelDownloadSnapshot = {
+  isDownloading: false,
+  progress: 0,
+  speed: 0,
+  downloaded: 0,
+  total: 0,
+  activeFile: null,
+  error: null,
+};
+
+function toActiveFile(repo: string, fileName: string): string {
+  return `${repo}:${fileName}`;
+}
+
+export function useModelDownload(token: string): UseModelDownloadState {
+  const [state, setState] = useState<ModelDownloadSnapshot>(INITIAL_STATE);
+  const [didComplete, setDidComplete] = useState(false);
+  const [downloadedFilePath, setDownloadedFilePath] = useState<string | null>(null);
+  const activeDownloadIdRef = useRef<string | null>(null);
+  const tokenRef = useRef(token);
+
+  useEffect(() => {
+    tokenRef.current = token;
+  }, [token]);
+
+  useEffect(() => {
+    let mounted = true;
+
+    const unsubscribePromise = listen<HuggingFaceDownloadProgress>(HUGGING_FACE_DOWNLOAD_EVENT, (event) => {
+      const payload = event.payload;
+      if (payload.downloadId !== activeDownloadIdRef.current) {
+        return;
+      }
+
+      if (!mounted) {
+        return;
+      }
+
+      const safeProgress = Number.isFinite(payload.progressPercentage) ? payload.progressPercentage : 0;
+      const safeSpeed = Number.isFinite(payload.downloadSpeed) ? payload.downloadSpeed : 0;
+      const safeDownloaded = Number.isFinite(payload.downloadedMB) ? payload.downloadedMB : 0;
+      const safeTotal = payload.totalMB != null && Number.isFinite(payload.totalMB) ? payload.totalMB : 0;
+
+      const isActive = payload.phase === "started" || payload.phase === "progress";
+      const isComplete = payload.phase === "completed";
+
+      setState((current) => ({
+        ...current,
+        isDownloading: isActive,
+        progress: safeProgress,
+        speed: safeSpeed,
+        downloaded: safeDownloaded,
+        total: safeTotal,
+        activeFile: toActiveFile(payload.repo, payload.fileName),
+        error: payload.error ?? null,
+      }));
+
+      if (isComplete) {
+        const finalPath = payload.destinationPath ?? null;
+        console.log(
+          "[HuggingFace] Download concluído! Arquivo em:",
+          finalPath
+        );
+        setDownloadedFilePath(finalPath);
+        setDidComplete(true);
+        activeDownloadIdRef.current = null;
+      } else if (payload.phase === "cancelled" || payload.phase === "error") {
+        activeDownloadIdRef.current = null;
+      }
+    });
+
+    return () => {
+      mounted = false;
+      void unsubscribePromise.then((unlisten) => unlisten());
+    };
+  }, []);
+
+  const startDownload = useCallback(async (repo: string, fileName: string) => {
+    try {
+      const downloadId = await startHuggingFaceDownload(repo, fileName, tokenRef.current);
+      activeDownloadIdRef.current = downloadId;
+      setState({
+        isDownloading: true,
+        progress: 0,
+        speed: 0,
+        downloaded: 0,
+        total: 0,
+        activeFile: toActiveFile(repo, fileName),
+        error: null,
+      });
+      console.log("[HuggingFace] Download iniciado:", repo, fileName);
+    } catch (err) {
+      console.error("[useModelDownload] startDownload failed:", err);
+      setState({
+        isDownloading: false,
+        progress: 0,
+        speed: 0,
+        downloaded: 0,
+        total: 0,
+        activeFile: null,
+        error: err instanceof Error ? err.message : String(err),
+      });
+      activeDownloadIdRef.current = null;
+    }
+  }, []);
+
+  const cancelDownload = useCallback(async () => {
+    if (!activeDownloadIdRef.current) {
+      return;
+    }
+
+    try {
+      await cancelHuggingFaceDownload(activeDownloadIdRef.current);
+    } catch (err) {
+      console.error("[useModelDownload] cancelDownload failed:", err);
+    }
+  }, []);
+
+  return {
+    ...state,
+    didComplete,
+    downloadedFilePath,
+    resetDidComplete: () => {
+      setDidComplete(false);
+      setDownloadedFilePath(null);
+    },
+    startDownload,
+    cancelDownload,
+  };
+}

tauri-app/src/features/settings/SettingsPage.css

@@ -0,0 +1,801 @@
+/* ── Shell ── */
+.sp-shell {
+  display: flex;
+  flex: 1;
+  width: 100%;
+  min-width: 0;
+  background: var(--bg);
+  color: var(--text);
+  height: 100vh;
+  overflow: hidden;
+}
+
+/* ── Left Navigation ── */
+.sp-nav {
+  width: 220px;
+  min-width: 220px;
+  background: var(--manus-surface);
+  border-right: 1px solid var(--manus-elevated);
+  display: flex;
+  flex-direction: column;
+  height: 100vh;
+  overflow: hidden;
+}
+
+.sp-nav-header {
+  padding: 24px 16px 16px;
+  border-bottom: 1px solid var(--border);
+  flex-shrink: 0;
+}
+
+.sp-nav-brand {
+  font-size: 11px;
+  font-weight: 600;
+  color: var(--text3);
+  letter-spacing: 0.02em;
+  margin-bottom: 2px;
+}
+
+.sp-nav-title {
+  font-size: 20px;
+  font-weight: 600;
+  color: var(--text);
+  letter-spacing: -0.02em;
+}
+
+.sp-nav-items {
+  flex: 1;
+  overflow-y: auto;
+  padding: 8px;
+}
+
+.sp-nav-item {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  padding: 8px 12px;
+  font-size: 13px;
+  font-weight: 500;
+  color: var(--text2);
+  background: none;
+  border: none;
+  cursor: pointer;
+  border-radius: 8px;
+  width: 100%;
+  transition: background 0.12s, color 0.12s;
+}
+
+.sp-nav-item:hover {
+  background: var(--manus-elevated);
+  color: var(--manus-tx);
+}
+
+.sp-nav-item--active {
+  background: var(--manus-elevated);
+  color: var(--manus-tx);
+  font-weight: 500;
+}
+
+/* Runtime dot in nav */
+.sp-nav-status {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  padding: 10px 16px;
+  border-top: 1px solid var(--border);
+  flex-shrink: 0;
+}
+
+.sp-nav-status-label {
+  color: var(--text3);
+}
+
+.sp-nav-footer {
+  padding: 12px 16px;
+  border-top: 1px solid var(--border);
+  flex-shrink: 0;
+}
+
+.sp-back-btn {
+  background: none;
+  border: none;
+  color: var(--text3);
+  font-size: 13px;
+  font-family: inherit;
+  cursor: pointer;
+  padding: 0;
+  transition: color 0.12s;
+}
+
+.sp-back-btn:hover {
+  color: var(--text);
+}
+
+/* ── Status dot ── */
+.sp-dot {
+  display: inline-block;
+  width: 7px;
+  height: 7px;
+  border-radius: 50%;
+  flex-shrink: 0;
+}
+
+.sp-dot--inline {
+  width: 6px;
+  height: 6px;
+  background-color: currentColor;
+  box-shadow: 0 0 6px currentColor;
+}
+
+.sp-dot--success { background: var(--success); box-shadow: 0 0 6px rgba(48,209,88,0.55); }
+.sp-dot--warn    { background: var(--warning);  animation: sp-pulse 1.2s ease-in-out infinite; }
+.sp-dot--danger  { background: var(--danger); }
+.sp-dot--idle    { background: var(--text3); }
+
+@keyframes sp-pulse {
+  0%, 100% { opacity: 1; }
+  50%       { opacity: 0.35; }
+}
+
+/* ── Main area (right side of sidebar) ── */
+.settings-main-area {
+  flex: 1;
+  display: flex;
+  flex-direction: column;
+  position: relative;
+  height: 100vh;
+  background: var(--bg-main, #0B0E14);
+  overflow: hidden;
+}
+
+.settings-scroll-container {
+  flex: 1;
+  overflow-y: auto;
+  width: 100%;
+}
+
+.settings-content {
+  width: 100%;
+  max-width: 800px;
+  margin: 0 auto;
+  padding: 48px 32px 120px 32px;
+}
+
+.page-header {
+  margin-bottom: 28px;
+}
+
+.settings-cards {
+  display: flex;
+  flex-direction: column;
+  gap: 24px;
+}
+
+.sp-section-header {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 16px;
+  margin-bottom: 4px;
+}
+
+.sp-section-title {
+  font-size: 24px;
+  font-weight: 600;
+  color: var(--manus-tx);
+  letter-spacing: -0.02em;
+  margin-bottom: 4px;
+}
+
+.sp-section-desc {
+  font-size: 14px;
+  color: var(--manus-tx2);
+  line-height: 1.5;
+}
+
+/* ── Status pill ── */
+.sp-status-pill {
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+  padding: 4px 10px;
+  border-radius: 9999px;
+  font-size: 11px;
+  font-weight: 600;
+  text-transform: uppercase;
+  letter-spacing: 0.05em;
+  background: var(--bg3);
+  color: var(--text2);
+  border: 1px solid var(--border);
+  flex-shrink: 0;
+}
+
+.sp-status-pill--success {
+  background: rgba(48,209,88,0.08);
+  color: var(--success);
+  border-color: rgba(48,209,88,0.2);
+}
+
+.sp-status-pill--warn {
+  background: rgba(255,214,10,0.08);
+  color: var(--warning);
+  border-color: rgba(255,214,10,0.2);
+}
+
+.sp-status-pill--danger {
+  background: rgba(255,69,58,0.08);
+  color: var(--danger);
+  border-color: rgba(255,69,58,0.2);
+}
+
+/* ── Card (continuous, macOS/Linear style) ── */
+.sp-card {
+  background: var(--manus-elevated);
+  border: 1px solid var(--manus-hover);
+  border-radius: 12px;
+  overflow: hidden;
+}
+
+.sp-card-row {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 24px;
+  padding: 16px 20px;
+}
+
+.sp-card-row--full {
+  flex-direction: column;
+  align-items: stretch;
+  width: 100%;
+  gap: 12px;
+}
+
+.sp-card-row-label {
+  min-width: 0;
+  flex-shrink: 0;
+  width: 180px;
+  padding-top: 2px;
+}
+
+.sp-card-row-title {
+  font-size: 14px;
+  font-weight: 500;
+  color: var(--text);
+  margin-bottom: 2px;
+}
+
+.sp-card-row-desc {
+  font-size: 12px;
+  color: var(--text3);
+  line-height: 1.4;
+}
+
+.sp-card-row-control {
+  flex: 1;
+  min-width: 0;
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  align-items: flex-end;
+}
+
+.sp-card-row-control--wide {
+  align-items: stretch;
+}
+
+.sp-card-divider {
+  height: 1px;
+  background: var(--border);
+  margin: 0 20px;
+}
+
+/* ── Active model ── */
+.sp-model-name {
+  font-size: 14px;
+  font-weight: 600;
+  color: var(--text);
+  margin-bottom: 2px;
+}
+
+.sp-model-path {
+  font-size: 11px;
+  color: var(--text3);
+  font-family: "Cascadia Code", "JetBrains Mono", monospace;
+  word-break: break-all;
+  line-height: 1.4;
+}
+
+.sp-pending {
+  font-size: 12px;
+  color: var(--accent);
+  background: rgba(72,166,255,0.07);
+  border: 1px solid rgba(72,166,255,0.18);
+  border-radius: 6px;
+  padding: 6px 10px;
+}
+
+.sp-empty {
+  font-size: 13px;
+  color: var(--text3);
+  font-style: italic;
+}
+
+/* ── Folder path ── */
+.sp-folder-actions {
+  display: flex;
+  gap: 8px;
+  align-items: center;
+}
+
+.sp-folder-path {
+  font-size: 12px;
+  color: var(--text2);
+  font-family: "Cascadia Code", "JetBrains Mono", monospace;
+  background: var(--bg3);
+  padding: 6px 10px;
+  border-radius: 6px;
+  border: 1px solid var(--border);
+  word-break: break-all;
+  margin-top: 6px;
+}
+
+/* ── Model list ── */
+.sp-model-list {
+  border: 1px solid var(--border);
+  border-radius: 8px;
+  overflow: hidden;
+  width: 100%;
+}
+
+.sp-model-row {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  width: 100%;
+  padding: 10px 14px;
+  background: var(--bg3);
+  border: none;
+  border-bottom: 1px solid var(--border);
+  cursor: pointer;
+  transition: background 0.1s;
+  text-align: left;
+  color: var(--text);
+  font-family: inherit;
+  font-size: 13px;
+}
+
+.sp-model-row:last-child { border-bottom: none; }
+
+.sp-model-row:hover:not(:disabled) { background: var(--bg4); }
+
+.sp-model-row--selected {
+  background: rgba(72,166,255,0.07);
+}
+
+.sp-model-row--disabled { opacity: 0.45; pointer-events: none; }
+
+.sp-model-row-select {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  flex: 1;
+  min-width: 0;
+  background: none;
+  border: none;
+  cursor: pointer;
+  padding: 0;
+  font-family: inherit;
+  color: inherit;
+  text-align: left;
+}
+.sp-model-row-select:disabled { cursor: not-allowed; }
+
+.sp-model-row-name { font-size: 13px; color: var(--text); }
+
+.sp-remove-btn {
+  background: none;
+  border: none;
+  color: var(--text3);
+  cursor: pointer;
+  padding: 4px;
+  border-radius: 4px;
+  flex-shrink: 0;
+  font-family: inherit;
+  transition: background 0.12s, color 0.12s;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+}
+
+.sp-remove-btn:hover {
+  background: rgba(240,64,64,0.15);
+  color: var(--danger);
+}
+
+.sp-scanning-row {
+  padding: 12px 14px;
+  color: var(--text3);
+  font-size: 13px;
+  background: var(--bg3);
+  animation: sp-pulse 1.2s ease-in-out infinite;
+}
+
+/* ── Radio controls ── */
+.sp-radio {
+  display: inline-block;
+  width: 14px;
+  height: 14px;
+  border-radius: 50%;
+  border: 2px solid var(--border2);
+  flex-shrink: 0;
+  transition: border-color 0.12s, background 0.12s, box-shadow 0.12s;
+}
+
+.sp-radio--on {
+  border-color: var(--accent);
+  background: var(--accent);
+  box-shadow: 0 0 0 3px rgba(72,166,255,0.18);
+}
+
+/* ── Segmented control (macOS style) ── */
+.sp-segmented {
+  display: inline-flex;
+  background: var(--bg3);
+  border: 1px solid var(--border);
+  border-radius: 8px;
+  padding: 3px;
+  gap: 2px;
+}
+
+.sp-segmented-btn {
+  padding: 6px 14px;
+  border-radius: 6px;
+  font-size: 12px;
+  font-weight: 500;
+  cursor: pointer;
+  background: transparent;
+  border: none;
+  color: var(--text3);
+  font-family: inherit;
+  transition: background 0.12s, color 0.12s;
+  white-space: nowrap;
+}
+
+.sp-segmented-btn:hover:not(:disabled) {
+  color: var(--text);
+  background: var(--bg4);
+}
+
+.sp-segmented-btn--active {
+  background: var(--bg4);
+  color: var(--text);
+  font-weight: 500;
+  box-shadow: 0 1px 3px rgba(0,0,0,0.2);
+}
+
+.sp-segmented-btn:disabled { opacity: 0.45; cursor: not-allowed; }
+
+/* ── Context custom input ── */
+.sp-ctx-custom {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  margin-top: 4px;
+}
+
+/* ── Toggle switches (iOS style) ── */
+.sp-toggles {
+  display: flex;
+  gap: 16px;
+}
+
+.sp-toggle {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  cursor: pointer;
+  font-size: 13px;
+  color: var(--text2);
+  transition: color 0.15s;
+}
+
+.sp-toggle input[type="checkbox"] {
+  display: none;
+}
+
+.sp-toggle-track {
+  width: 36px;
+  height: 20px;
+  background: var(--bg4);
+  border: 1px solid var(--border);
+  border-radius: 10px;
+  position: relative;
+  transition: background 0.2s, border-color 0.2s;
+  flex-shrink: 0;
+}
+
+.sp-toggle-track::after {
+  content: "";
+  position: absolute;
+  top: 2px;
+  left: 2px;
+  width: 14px;
+  height: 14px;
+  background: var(--text3);
+  border-radius: 50%;
+  transition: transform 0.2s, background 0.2s;
+}
+
+.sp-toggle--active .sp-toggle-track {
+  background: var(--accent);
+  border-color: var(--accent);
+}
+
+.sp-toggle--active .sp-toggle-track::after {
+  transform: translateX(16px);
+  background: #fff;
+}
+
+.sp-toggle--active {
+  color: var(--text);
+}
+
+.sp-toggle-label {
+  font-weight: 500;
+}
+
+.sp-toggle-desc {
+  font-size: 12px;
+  color: var(--text3);
+  margin-top: 4px;
+  line-height: 1.4;
+}
+
+/* ── GPU ── */
+.sp-gpu-info-row {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+  padding: 10px 12px;
+  background: var(--bg3);
+  border-radius: 8px;
+  border: 1px solid var(--border);
+}
+
+.sp-gpu-name  { font-size: 13px; font-weight: 600; color: var(--text); }
+.sp-gpu-vram  { font-size: 12px; color: var(--text3); font-family: monospace; }
+
+/* ── Slider ── */
+.sp-slider {
+  -webkit-appearance: none;
+  appearance: none;
+  height: 4px;
+  background: var(--bg5);
+  border-radius: 2px;
+  outline: none;
+  cursor: pointer;
+  display: block;
+}
+
+.sp-slider::-webkit-slider-thumb {
+  -webkit-appearance: none;
+  width: 16px;
+  height: 16px;
+  border-radius: 50%;
+  background: var(--accent);
+  cursor: pointer;
+  box-shadow: 0 0 5px rgba(72,166,255,0.4);
+  transition: box-shadow 0.15s;
+}
+
+.sp-slider::-webkit-slider-thumb:hover {
+  box-shadow: 0 0 0 4px rgba(72,166,255,0.2);
+}
+
+.sp-slider-presets {
+  display: flex;
+  gap: 12px;
+  margin-top: 4px;
+}
+
+/* ── Field label ── */
+.sp-field-header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+}
+
+.sp-field-label {
+  font-size: 12px;
+  color: var(--text3);
+  font-weight: 600;
+}
+
+.sp-field-value {
+  font-size: 12px;
+  font-weight: 600;
+  color: var(--text2);
+  font-family: "Cascadia Code", "JetBrains Mono", monospace;
+}
+
+/* ── Inputs ── */
+.sp-input {
+  background: var(--manus-surface);
+  border: 1px solid var(--manus-elevated);
+  border-radius: 8px;
+  color: var(--manus-tx);
+  font-size: 13px;
+  padding: 8px 12px;
+  outline: none;
+  font-family: inherit;
+  transition: border-color 0.15s, background 0.15s;
+}
+
+.sp-input:hover {
+  border-color: var(--manus-hover);
+}
+
+.sp-input:focus {
+  border-color: var(--manus-blue);
+  background: var(--manus-elevated);
+  box-shadow: 0 0 0 1px var(--manus-blue);
+}
+
+.sp-input:disabled {
+  opacity: 0.45;
+  cursor: not-allowed;
+}
+
+.sp-input--sm {
+  width: 120px;
+  padding: 6px 10px;
+  font-size: 12px;
+}
+
+/* ── Hint text ── */
+.sp-hint {
+  font-size: 12px;
+  color: var(--text3);
+  line-height: 1.5;
+}
+
+.sp-hint code {
+  font-family: "Cascadia Code", "JetBrains Mono", monospace;
+  font-size: 11px;
+  color: var(--text2);
+  background: var(--bg4);
+  padding: 1px 5px;
+  border-radius: 3px;
+}
+
+/* ── Buttons ── */
+.sp-btn {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 6px;
+  padding: 8px 16px;
+  border-radius: 8px;
+  font-size: 13px;
+  font-weight: 500;
+  font-family: inherit;
+  cursor: pointer;
+  border: 1px solid transparent;
+  transition: background 0.12s, color 0.12s, border-color 0.12s, opacity 0.12s;
+  white-space: nowrap;
+  line-height: 1;
+}
+
+.sp-btn:disabled { opacity: 0.4; cursor: not-allowed; }
+
+.sp-btn--primary {
+  background: var(--text);
+  color: var(--bg);
+  border-color: var(--text);
+}
+.sp-btn--primary:hover:not(:disabled) { opacity: 0.85; }
+
+.sp-btn--secondary {
+  background: var(--bg3);
+  color: var(--text2);
+  border-color: var(--border);
+}
+.sp-btn--secondary:hover:not(:disabled) { background: var(--bg4); color: var(--text); }
+
+.sp-btn--ghost {
+  background: transparent;
+  color: var(--text3);
+  border-color: transparent;
+}
+.sp-btn--ghost:hover:not(:disabled) { background: var(--bg3); color: var(--text); }
+
+.sp-btn--danger {
+  background: transparent;
+  color: var(--danger);
+  border-color: rgba(240,64,64,0.3);
+}
+.sp-btn--danger:hover:not(:disabled) { background: rgba(240,64,64,0.1); }
+
+.sp-btn-xs {
+  font-size: 11px;
+  color: var(--text3);
+  background: none;
+  border: none;
+  cursor: pointer;
+  padding: 2px 6px;
+  font-family: inherit;
+  transition: color 0.12s;
+  border-radius: 4px;
+}
+.sp-btn-xs:hover { color: var(--accent); }
+
+/* ── Floating action bar ── */
+.settings-action-bar {
+  position: absolute;
+  bottom: 0;
+  left: 0;
+  width: 100%;
+  background: rgba(11, 14, 20, 0.85);
+  backdrop-filter: blur(12px);
+  -webkit-backdrop-filter: blur(12px);
+  border-top: 1px solid rgba(255, 255, 255, 0.05);
+  z-index: 50;
+}
+
+.action-bar-inner {
+  width: 100%;
+  max-width: 800px;
+  margin: 0 auto;
+  padding: 16px 32px;
+  display: flex;
+  justify-content: flex-end;
+  align-items: center;
+  gap: 12px;
+}
+
+.sp-error-inline {
+  font-size: 12px;
+  color: var(--danger);
+  background: rgba(240,64,64,0.08);
+  border: 1px solid rgba(240,64,64,0.2);
+  border-radius: 6px;
+  padding: 6px 12px;
+  line-height: 1.4;
+}
+
+/* ── Diagnostics ── */
+.sp-diag-list {
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+  width: 100%;
+}
+
+.sp-diag-row {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 12px;
+  font-size: 13px;
+}
+
+.sp-diag-key { color: var(--text3); flex-shrink: 0; }
+.sp-diag-val { color: var(--text2); text-align: right; word-break: break-all; }
+
+.sp-mono   { font-family: "Cascadia Code", "JetBrains Mono", monospace; font-size: 11px; }
+.sp-accent { color: var(--accent); }
+.sp-danger { color: var(--danger); }
+
+/* ── Utilities ── */
+.mt-2 { margin-top: 8px; }
+.mt-3 { margin-top: 12px; }
+.w-full { width: 100%; }
+.w-32 { width: 8rem; }
+.flex { display: flex; }
+.flex-1 { flex: 1; }
+.items-center { align-items: center; }
+.gap-2 { gap: 8px; }
+.gap-3 { gap: 12px; }
+.truncate { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }

tauri-app/src/features/settings/SettingsPage.download.test.tsx

@@ -0,0 +1,161 @@
+/* @vitest-environment jsdom */
+
+import { act } from "react";
+import { createRoot, type Root } from "react-dom/client";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+import SettingsPage from "./SettingsPage";
+import type { RuntimeSnapshot } from "../runtime/types";
+
+const startDownloadMock = vi.fn();
+const cancelDownloadMock = vi.fn();
+
+const modelDownloadState = {
+  isDownloading: false,
+  progress: 0,
+  speed: 0,
+  downloaded: 0,
+  total: 0,
+  activeFile: null as string | null,
+  error: null as string | null,
+  startDownload: startDownloadMock,
+  cancelDownload: cancelDownloadMock,
+};
+
+vi.mock("../runtime/api", () => ({
+  pickRuntimeModelFiles: vi.fn(async () => []),
+  pickRuntimeModelsDir: vi.fn(async () => null),
+  queryRuntimeGpu: vi.fn(async () => null),
+  searchHuggingFaceModels: vi.fn(async () => [
+    {
+      repo_id: "microsoft/phi-4",
+      author: "microsoft",
+      verified_files: [
+        {
+          path: "quantized/phi-4-q4.gguf",
+          size: 4_294_967_296,
+          quantization: "q4",
+        },
+      ],
+    },
+  ]),
+}));
+
+vi.mock("../runtime/useModelDownload", () => ({
+  useModelDownload: () => modelDownloadState,
+}));
+
+function createSnapshot(): RuntimeSnapshot {
+  return {
+    status: "stopped",
+    ready: false,
+    process_running: false,
+    resolved_url: "",
+    binary_path: null,
+    last_error: null,
+    active_command: "idle",
+    last_command_result: {
+      command: "save",
+      outcome: "succeeded",
+      message: "",
+    },
+    config: {
+      connection_mode: "local",
+      start_policy: "manual",
+      server_url: "",
+      api_key: "",
+      model_path: null,
+      models_dir: null,
+      addr: "127.0.0.1:8080",
+      gpu_layers: 0,
+      ctx_size: 4096,
+      extra_model_paths: [],
+      selected_model_id: null,
+      remote_presets: [],
+      hugging_face_token: "hf_token",
+    },
+  };
+}
+
+function setInputValue(input: HTMLInputElement, value: string) {
+  const descriptor = Object.getOwnPropertyDescriptor(HTMLInputElement.prototype, "value");
+  descriptor?.set?.call(input, value);
+  input.dispatchEvent(new Event("input", { bubbles: true }));
+  input.dispatchEvent(new Event("change", { bubbles: true }));
+}
+
+describe("SettingsPage Hugging Face downloads", () => {
+  let container: HTMLDivElement;
+  let root: Root;
+
+  beforeEach(() => {
+    // @ts-expect-error React test environment flag
+    globalThis.IS_REACT_ACT_ENVIRONMENT = true;
+    container = document.createElement("div");
+    document.body.appendChild(container);
+    root = createRoot(container);
+    startDownloadMock.mockReset();
+    cancelDownloadMock.mockReset();
+    Object.assign(modelDownloadState, {
+      isDownloading: false,
+      progress: 0,
+      speed: 0,
+      downloaded: 0,
+      total: 0,
+      activeFile: null,
+      error: null,
+    });
+  });
+
+  afterEach(() => {
+    act(() => root.unmount());
+    container.remove();
+    // @ts-expect-error React test environment flag
+    globalThis.IS_REACT_ACT_ENVIRONMENT = false;
+  });
+
+  it("starts a real download from the Hugging Face row", async () => {
+    const runtime = {
+      snapshot: createSnapshot(),
+      savedBaseline: createSnapshot().config,
+      draftConfig: createSnapshot().config,
+      draftDirty: false,
+      loading: false,
+      busy: false,
+      lifecycleBusy: false,
+      scannedModels: [],
+      isScanningModels: false,
+      updateDraft: vi.fn(),
+      saveConfig: vi.fn(),
+      startLocal: vi.fn(),
+      stopLocal: vi.fn(),
+      scanModels: vi.fn(async () => undefined),
+      resetDraft: vi.fn(),
+    };
+
+    await act(async () => {
+      root.render(<SettingsPage runtime={runtime} onClose={() => undefined} />);
+    });
+
+    await act(async () => {
+      const input = container.querySelector('input[placeholder="Search for GGUF models…"]') as HTMLInputElement;
+      setInputValue(input, "phi-4");
+      await Promise.resolve();
+    });
+
+    await act(async () => {
+      (Array.from(container.querySelectorAll("button")).find((button) => button.textContent?.includes("Search")) as HTMLButtonElement).click();
+      await Promise.resolve();
+      await Promise.resolve();
+    });
+
+    const fileRow = container.querySelector(".sp-hf-file");
+    expect(fileRow).not.toBeNull();
+
+    await act(async () => {
+      (fileRow?.querySelector('button[aria-label^="Download "]') as HTMLButtonElement).click();
+      await Promise.resolve();
+    });
+
+    expect(startDownloadMock).toHaveBeenCalledWith("microsoft/phi-4", "quantized/phi-4-q4.gguf");
+  });
+});

tauri-app/src/index.css

@@ -2,41 +2,57 @@
 *, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
 
 :root {
-  --bg:        #0a1016;
-  --bg2:       #0f1721;
-  --bg3:       #14202d;
-  --bg4:       #182636;
-  --bg5:       #203246;
-  --border:    #243241;
-  --border2:   #344759;
-  --text:      #edf5ff;
-  --text2:     #93a8bc;
-  --text3:     #5d7288;
-  --accent:    #48a6ff;
-  --accent2:   #2f8ae3;
-  --accent-glow: rgba(72,166,255,0.16);
-  --user-bg:   #10263f;
-  --user-border: #214468;
-  --asst-bg:   #101924;
-  --asst-border: #243241;
+  /* Updated to Manus design tokens */
+  --bg:        #171717;   /* Manus bg-base */
+  --bg2:       #1E1E1E;   /* Manus bg-surface */
+  --bg3:       #262626;   /* Manus bg-elevated */
+  --bg4:       #333333;   /* Manus bg-hover */
+  --bg5:       #3D3D3D;   /* slightly lighter */
+  --border:    #232a33;
+  --border2:   #313944;
+  --text:      #F3F4F6;   /* Manus text-primary */
+  --text2:     #9CA3AF;   /* Manus text-secondary */
+  --text3:     #52525B;   /* Manus text-muted */
+  --accent:    #4b86ff;
+  --accent2:   #3674ef;
+  --accent-glow: rgba(75,134,255,0.14);
+  --user-bg:   #101c2d;
+  --user-border: #203451;
+  --asst-bg:   #10161d;
+  --asst-border: #26303b;
   --danger:    #f04040;
   --success:   #30d158;
   --warning:   #ffd60a;
   --code-bg:   #0d0d10;
   --radius-sm: 6px;
+  --radius-md: 10px;
   --radius:    10px;
   --radius-lg: 14px;
+  --font-sans: "Inter", "Segoe UI Variable", "SF Pro Display", system-ui, sans-serif;
+  --font-serif: "Merriweather", Georgia, serif;
   --shadow:    0 2px 12px rgba(0,0,0,0.4);
   --shadow-lg: 0 8px 32px rgba(0,0,0,0.6);
   font-size: 15px;
 }
 
+/* Manus design tokens as named vars (for new components) */
+:root {
+  --manus-base:     #171717;
+  --manus-surface:  #1E1E1E;
+  --manus-elevated: #262626;
+  --manus-hover:    #333333;
+  --manus-tx:       #F3F4F6;
+  --manus-tx2:      #9CA3AF;
+  --manus-tx3:      #52525B;
+  --manus-blue:     #3B82F6;
+}
+
 html, body, #root {
   height: 100%;
   overflow: hidden;
   background: var(--bg);
   color: var(--text);
-  font-family: "Segoe UI Variable", "SF Pro Display", "IBM Plex Sans", "Inter", system-ui, sans-serif;
+  font-family: var(--font-sans);
   -webkit-font-smoothing: antialiased;
 }
 
@@ -54,3 +70,6 @@ html, body, #root {
 
 /* ── Highlight.js theme override ── */
 .hljs { background: var(--code-bg) !important; }
+
+/* ── Tailwind utilities (preflight disabled — existing styles preserved) ── */
+@tailwind utilities;

tauri-app/tailwind.config.js

@@ -0,0 +1,35 @@
+/** @type {import('tailwindcss').Config} */
+export default {
+  content: ["./index.html", "./src/**/*.{ts,tsx}"],
+  corePlugins: { preflight: false }, // preserve existing hand-crafted CSS
+  theme: {
+    extend: {
+      colors: {
+        "c-bg":      "var(--bg)",
+        "c-bg2":     "var(--bg2)",
+        "c-bg3":     "var(--bg3)",
+        "c-bg4":     "var(--bg4)",
+        "c-bg5":     "var(--bg5)",
+        "c-bdr":     "var(--border)",
+        "c-bdr2":    "var(--border2)",
+        "c-tx":      "var(--text)",
+        "c-tx2":     "var(--text2)",
+        "c-tx3":     "var(--text3)",
+        "c-ac":      "var(--accent)",
+        "c-ac2":     "var(--accent2)",
+        "c-danger":  "var(--danger)",
+        "c-success": "var(--success)",
+        "c-warn":    "var(--warning)",
+        // Manus design tokens
+        "manus-base":     "var(--manus-base)",
+        "manus-surface":  "var(--manus-surface)",
+        "manus-elevated": "var(--manus-elevated)",
+        "manus-hover":    "var(--manus-hover)",
+        "manus-tx":       "var(--manus-tx)",
+        "manus-tx2":      "var(--manus-tx2)",
+        "manus-tx3":      "var(--manus-tx3)",
+        "manus-blue":     "var(--manus-blue)",
+      },
+    },
+  },
+};

vendor/llama-runtime/manifest.json

@@ -0,0 +1,48 @@
+{
+  "release": "b8358",
+  "variants": [
+    {
+      "id": "windows-x64-cpu",
+      "target": "windows-x86_64",
+      "gpu": false,
+      "dll": "llama.dll",
+      "archives": [
+        {
+          "url": "https://github.com/ggml-org/llama.cpp/releases/download/b8358/llama-b8358-bin-win-cpu-x64.zip",
+          "sha256": "bb78ff8d7e9b53c36fe4cc845f8a479017f97bf5fd99a9a047b45c9c36b93810",
+          "size": 37300696
+        }
+      ],
+      "required_files": [
+        "llama.dll",
+        "ggml.dll",
+        "ggml-base.dll"
+      ]
+    },
+    {
+      "id": "windows-x64-cuda-12.4",
+      "target": "windows-x86_64",
+      "gpu": true,
+      "dll": "llama.dll",
+      "archives": [
+        {
+          "url": "https://github.com/ggml-org/llama.cpp/releases/download/b8358/llama-b8358-bin-win-cuda-12.4-x64.zip",
+          "sha256": "92c0c30401e7284fd28836b112ffa752b3beb7d89786e0c87e378f478b973e48",
+          "size": 226836797
+        },
+        {
+          "url": "https://github.com/ggml-org/llama.cpp/releases/download/b8358/cudart-llama-bin-win-cuda-12.4-x64.zip",
+          "sha256": "8c79a9b226de4b3cacfd1f83d24f962d0773be79f1e7b75c6af4ded7e32ae1d6",
+          "size": 391443627
+        }
+      ],
+      "required_files": [
+        "llama.dll",
+        "ggml.dll",
+        "ggml-base.dll",
+        "ggml-cuda.dll",
+        "cudart64_12.dll"
+      ]
+    }
+  ]
+}