fix: remove missing resource path from tauri.conf; add GPU detection and layer clamping

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

@@ -4,6 +4,57 @@ use tauri::{Manager, State};
 
 pub struct ServerState(pub Mutex<Option<Child>>);
 
+/// Query GPU info via nvidia-smi.
+/// Returns (gpu_name, vram_mib, recommended_layers).
+/// recommended_layers is estimated as vram_mib / 80 (rough heuristic for 4-bit quant).
+#[tauri::command]
+fn query_gpu() -> Result<GpuInfo, String> {
+    let output = Command::new("nvidia-smi")
+        .args([
+            "--query-gpu=name,memory.total",
+            "--format=csv,noheader,nounits",
+        ])
+        .output()
+        .map_err(|_| "nvidia-smi not found — no NVIDIA GPU detected".to_string())?;
+
+    if !output.status.success() {
+        return Err("nvidia-smi failed — no NVIDIA GPU detected".into());
+    }
+
+    let stdout = String::from_utf8_lossy(&output.stdout);
+    let line = stdout.lines().next().unwrap_or("").trim().to_string();
+    if line.is_empty() {
+        return Err("No GPU found".into());
+    }
+
+    // Format: "GPU Name, 8192"
+    let parts: Vec<&str> = line.splitn(2, ',').collect();
+    if parts.len() < 2 {
+        return Err(format!("Unexpected nvidia-smi output: {line}"));
+    }
+
+    let name = parts[0].trim().to_string();
+    let vram_mib: u32 = parts[1].trim().parse().unwrap_or(0);
+
+    // Heuristic: ~80 MiB per layer for Q4/Q6 models (conservative)
+    // Reserve 1 GiB for OS/other usage
+    let usable = vram_mib.saturating_sub(1024);
+    let recommended_layers = (usable / 80).min(200) as i32;
+
+    Ok(GpuInfo {
+        name,
+        vram_mib,
+        recommended_layers,
+    })
+}
+
+#[derive(serde::Serialize)]
+pub struct GpuInfo {
+    pub name: String,
+    pub vram_mib: u32,
+    pub recommended_layers: i32,
+}
+
 #[tauri::command]
 fn start_server(
     state: State<ServerState>,
@@ -17,7 +68,6 @@ fn start_server(
         return Err("Server is already running".into());
     }
 
-    // Resolve binary path relative to the app's resource dir or PATH
     let binary = if cfg!(target_os = "windows") {
         "localllm.exe"
     } else {
@@ -70,6 +120,7 @@ pub fn run() {
         .plugin(tauri_plugin_shell::init())
         .manage(ServerState(Mutex::new(None)))
         .invoke_handler(tauri::generate_handler![
+            query_gpu,
             start_server,
             stop_server,
             server_status

tauri-app/src-tauri/tauri.conf.json

@@ -29,8 +29,7 @@
   "bundle": {
     "active": true,
     "targets": "all",
-    "icon": [],
-    "resources": ["../../../target/release/localllm.exe", "../../../target/release/localllm"]
+    "icon": []
   },
   "plugins": {
     "shell": {

tauri-app/src/components/ServerPanel.css

@@ -144,3 +144,90 @@
 .sp-message-error {
   color: #ef4444;
 }
+
+.sp-gpu-badge {
+  font-size: 11px;
+  color: #22c55e;
+  background: rgba(34, 197, 94, 0.08);
+  border: 1px solid rgba(34, 197, 94, 0.2);
+  border-radius: 5px;
+  padding: 4px 8px;
+  margin-top: 6px;
+}
+
+.sp-gpu-cpu {
+  color: #9ca3af;
+  background: rgba(156, 163, 175, 0.08);
+  border-color: rgba(156, 163, 175, 0.2);
+}
+
+.sp-gpu-loading {
+  color: #f59e0b;
+  background: rgba(245, 158, 11, 0.08);
+  border-color: rgba(245, 158, 11, 0.2);
+}
+
+.sp-label-hint {
+  color: var(--accent, #5b6af0);
+  font-weight: 400;
+}
+
+.sp-layers-row {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.sp-btn-max {
+  background: var(--accent, #5b6af0);
+  color: #fff;
+  border: none;
+  border-radius: 5px;
+  padding: 5px 10px;
+  font-size: 11px;
+  font-weight: 600;
+  cursor: pointer;
+  transition: opacity 0.15s;
+}
+
+.sp-btn-max:hover:not(:disabled) {
+  opacity: 0.85;
+}
+
+.sp-btn-max:disabled {
+  opacity: 0.4;
+  cursor: not-allowed;
+}
+
+.sp-layers-hint {
+  font-size: 11px;
+  color: #9ca3af;
+}
+
+.sp-vram-bar {
+  position: relative;
+  height: 18px;
+  background: rgba(255,255,255,0.06);
+  border-radius: 4px;
+  overflow: hidden;
+  margin-top: 4px;
+}
+
+.sp-vram-fill {
+  height: 100%;
+  background: linear-gradient(90deg, var(--accent, #5b6af0), #818cf8);
+  transition: width 0.3s ease;
+  border-radius: 4px;
+}
+
+.sp-vram-label {
+  position: absolute;
+  inset: 0;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  font-size: 10px;
+  color: #e5e7eb;
+  font-weight: 600;
+  text-shadow: 0 1px 2px rgba(0,0,0,0.5);
+}

tauri-app/src/components/ServerPanel.tsx

@@ -1,4 +1,5 @@
-import { useState } from "react";
+import { useState, useEffect } from "react";
+import { invoke } from "@tauri-apps/api/core";
 import type { ProcessStatus, ServerConfig } from "../hooks/useServerProcess";
 import "./ServerPanel.css";
 
@@ -9,6 +10,12 @@ interface Props {
   onStop: () => void;
 }
 
+interface GpuInfo {
+  name: string;
+  vram_mib: number;
+  recommended_layers: number;
+}
+
 const STATUS_LABEL: Record<ProcessStatus, string> = {
   stopped: "Stopped",
   starting: "Starting…",
@@ -35,9 +42,31 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
   const [gpuLayers, setGpuLayers] = useState(
     () => parseInt(localStorage.getItem("forgelm_gpu_layers") || "0", 10)
   );
+  const [gpuInfo, setGpuInfo] = useState<GpuInfo | null>(null);
+  const [gpuError, setGpuError] = useState<string>("");
   const [expanded, setExpanded] = useState(false);
 
+  // Detect GPU on mount
+  useEffect(() => {
+    invoke<GpuInfo>("query_gpu")
+      .then((info) => {
+        setGpuInfo(info);
+        // Only auto-set if user hasn't saved a preference yet
+        if (!localStorage.getItem("forgelm_gpu_layers")) {
+          setGpuLayers(info.recommended_layers);
+        }
+      })
+      .catch((e) => {
+        setGpuError(String(e));
+        // Default to CPU-only if no GPU
+        if (!localStorage.getItem("forgelm_gpu_layers")) {
+          setGpuLayers(0);
+        }
+      });
+  }, []);
+
   const isActive = status === "running" || status === "starting" || status === "stopping";
+  const maxLayers = gpuInfo ? gpuInfo.recommended_layers : 999;
 
   const handleStart = () => {
     localStorage.setItem("forgelm_model_path", modelPath);
@@ -46,6 +75,11 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
     onStart({ modelPath, addr, gpuLayers });
   };
 
+  const handleLayersChange = (val: number) => {
+    const clamped = Math.max(0, Math.min(val, maxLayers));
+    setGpuLayers(clamped);
+  };
+
   return (
     <div className="server-panel">
       <div className="server-panel-header" onClick={() => setExpanded(!expanded)}>
@@ -64,6 +98,19 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
 
       {expanded && (
         <div className="server-panel-body">
+          {/* GPU info badge */}
+          {gpuInfo ? (
+            <div className="sp-gpu-badge">
+              🎮 {gpuInfo.name} — {(gpuInfo.vram_mib / 1024).toFixed(1)} GB VRAM
+            </div>
+          ) : gpuError ? (
+            <div className="sp-gpu-badge sp-gpu-cpu">
+              💻 CPU only — {gpuError.includes("not found") ? "No NVIDIA GPU" : gpuError}
+            </div>
+          ) : (
+            <div className="sp-gpu-badge sp-gpu-loading">🔍 Detecting GPU…</div>
+          )}
+
           <label className="sp-label">Model path (.gguf)</label>
           <input
             className="sp-input"
@@ -84,16 +131,48 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
             disabled={isActive}
           />
 
-          <label className="sp-label">GPU Layers (0 = CPU only)</label>
-          <input
-            className="sp-input sp-input-num"
-            type="number"
-            min={0}
-            max={999}
-            value={gpuLayers}
-            onChange={(e) => setGpuLayers(parseInt(e.target.value, 10) || 0)}
-            disabled={isActive}
-          />
+          <label className="sp-label">
+            GPU Layers
+            {gpuInfo && (
+              <span className="sp-label-hint"> (max {gpuInfo.recommended_layers} for your GPU)</span>
+            )}
+          </label>
+          <div className="sp-layers-row">
+            <input
+              className="sp-input sp-input-num"
+              type="number"
+              min={0}
+              max={maxLayers}
+              value={gpuLayers}
+              onChange={(e) => handleLayersChange(parseInt(e.target.value, 10) || 0)}
+              disabled={isActive}
+            />
+            {gpuInfo && (
+              <button
+                className="sp-btn-max"
+                onClick={() => handleLayersChange(gpuInfo.recommended_layers)}
+                disabled={isActive}
+                title="Set to maximum recommended"
+              >
+                Max
+              </button>
+            )}
+            {gpuLayers === 0 && (
+              <span className="sp-layers-hint">CPU only</span>
+            )}
+          </div>
+
+          {gpuInfo && gpuLayers > 0 && (
+            <div className="sp-vram-bar">
+              <div
+                className="sp-vram-fill"
+                style={{ width: `${Math.min(100, (gpuLayers / gpuInfo.recommended_layers) * 100)}%` }}
+              />
+              <span className="sp-vram-label">
+                ~{Math.round(gpuLayers * 80)} MB / {gpuInfo.vram_mib} MB VRAM
+              </span>
+            </div>
+          )}
 
           <div className="sp-actions">
             {status === "stopped" || status === "error" ? (