Added the 'You are a helpful local ai assistant' pattern to the hasKnownPromptEcho matcher so the UI correctly strips transcript bleed before displaying or saving it to SQLite.
Added orchestrator.rs with parallel dispatching of 4 personas (Security, Architect, Critic, UX) using tokio Semaphores to prevent OOM. Included stream multiplexing and synthesis phase, though still in draft/skeleton form.
Added a catch-all parsing step at the very end of the streaming loop, and within the flushed chunk loop, so any straggler text that wasn't accompanied by a finish_reason chunk or was buffered in streamDecoder.flush() is properly parsed into visible content before the DB insert.
Added detailed analysis of 20 distinct failure points including Context Drifting, Synthesis Poisoning, OOM risks, UI freezes, Tokenization bottlenecks, and Attention Dilution, along with their respective architectural mitigations in Rust and React.
The redesign of the chat UI to use StreamingBubble mistakenly omitted the actual rendering of the visible 'content' (and fallback raw text). It only showed the thinking/reasoning block and the loading dots, leaving the main content invisible until the generation finished. Added the markdown rendering block back to StreamingBubble to show text in real-time.
When using local aliases like Cooper, Iron, or Steel (which are Qwen-based architectures), the select_chat_template failed to identify them because the alias names didn't contain 'qwen'. This caused the inference server to fall back to the Generic template (User: ... \nAssistant:), which caused Qwen models to hallucinate completion blocks starting with <|im_end|>\n<|im_start|>assistant to 'correct' the missing tags, resulting in the model repeating the user's prompt at the beginning of its response.
Passes the last_error.message from the runtime snapshot down through App to ChatView and into the chat debug formatter. This ensures that any specific errors that occur during local API startup, model loading, or operation are captured directly in the debug export.
Added 'runtimeMode' and 'runtimeStatus' to the 'Copy Debug' transcript. This helps diagnose model loading errors or API initialization issues when users export chat logs for troubleshooting.
Moved the useEffect that auto-scans models below the definition of scanModels to fix a ReferenceError caused by the Temporal Dead Zone (TDZ) when calling a block-scoped useCallback function before it was initialized.
1. Reduced repeat_penalty to 1.15 to avoid degrading normal generation.
2. TPS degradation monitor now actively aborts the stream instead of just warning.
3. Removed broad transcript leakage regex from the frontend, relying fully on backend stop sequences to prevent false positives.
4. Fixed race condition in useRuntimeManager by moving model scanning into a reactive useEffect instead of an imperative ref call.
5. Enforced JSON format natively via response_format for ForgeAgent inference requests.
When a model generates only a reasoning block (<think>...</think>) with no visible content, the reasoning was previously lost because the DB insert omitted the reasoning_content column while using a fallback error message.
This separates the hasReasoningOnly condition from the corrupted/transcript leakage conditions, ensuring the reasoning is properly saved and the user sees a specific placeholder.
Backend (inference-server/compat.rs):
- PreparedPrompt now carries extra_stop_sequences injected by template
- Generic template injects User:/System:/Assistant:/Human: stop sequences
to prevent the model from continuing the transcript after its reply
- Increased repeat_penalty 1.15→1.3 and repeat_last_n 256→512 to
break repetition loops more aggressively
Backend (inference-server/routes.rs):
- Merge template extra_stop_sequences into sampler before inference
Frontend (chat.ts):
- Added 'System:' and 'Human:' to CHAT_TRANSCRIPT_STOP_SEQUENCES
- Extended TRANSCRIPT_LEAKAGE_PATTERN to include 'Human:'
Frontend (ChatView.tsx):
- buildApiMessages now strips leaked transcript from assistant messages
before sending history back to the model
- Reduced history window to 20 messages to limit context pollution
- Removed complex ForgeAgent state management from ChatView and MessageBubble
- Messages now display as simple text with markdown rendering
- Removed useMemoryExtractor hook and memory system prompt logic
- Simplified message persistence (no forgeagent_state column)
- Fixed agentic mode to show AI response as text instead of fake plan
This restores the working chat flow where AI responses appear as text
in the interface, fixing the issue where 'servidor de inferência não
disponível' error was shown instead of the actual AI response.
- Check process_running and ready state before making inference request
- Emit descriptive error messages for each failure scenario
- Handle local vs remote connection modes appropriately
- Validation score: 100 (was 0)
Result: {"status":"keep","validation_score":100,"response_time_ms":500}
- Apply Manus design tokens to index.css and tailwind.config.js
- Replace three-dot thinking indicator with pulsing blue dot + 'Thinking' text
- Change send button icon from ArrowUp to PaperPlaneRight
- Align message and input widths to 800px max
- Fix Rust model scan to only list top-level models (not subdirectories)
- Add scan_model_dir_top_level() function in model_files.rs
- Create ModelSelectorPill component for chat input area
- Remove model-section from Sidebar, move to ChatView
- Update ChatView props to receive model-related props
- Update tests for new ChatView props
- routes.rs: StreamItem::Error now emits content:None so fullContent stays
empty and the frontend does not accumulate 'internal error' as a response
- ChatView.tsx: buildApiMessages filters out empty/error messages from the
history sent to the backend, preventing corrupted context on retries
- ChatView.tsx: assistant INSERT is guarded — skipped when fullContent is
empty, so a failed stream leaves no empty assistant row in the DB
- engine.rs: new_session() now returns Err when llama_get_memory returns NULL
instead of silently continuing, preventing decode corruption on 2nd request
- session.rs: decode_prefill error includes token count for easier diagnosis
