Add Ollama API integration

frontend/src/App.js

@@ -16,6 +16,7 @@ function App() {
   const [activeTab, setActiveTab] = useState('chat');
   const [settings, setSettings] = useState({
     model: 'gpt-4o-mini',
+    memoryModel: 'gpt-4o-mini',
     persona: 'helpful_assistant',
     timezone: 'America/New_York',
     serverUrl: 'http://localhost:47283'
@@ -57,7 +58,7 @@ function App() {
       if (response.ok) {
         const data = await response.json();
         console.log('API key status:', data);
-        
+
         if (forceOpen || (data.requires_api_key && data.missing_keys.length > 0)) {
           if (forceOpen) {
             console.log('Manual API key update requested');
@@ -88,10 +89,10 @@ function App() {
   // Refresh backend-dependent data after successful connection
   const refreshBackendData = useCallback(async () => {
     console.log('🔄 Refreshing backend-dependent data...');
-    
+
     // Check API keys after successful backend connection
     await checkApiKeys();
-    
+
     // Trigger refresh of other backend-dependent components
     // This will cause components like SettingsPanel to re-fetch their data
     setSettings(prev => ({ ...prev, lastBackendRefresh: Date.now() }));
@@ -101,7 +102,7 @@ function App() {
   const checkBackendHealth = useCallback(async () => {
     let shouldProceed = true;
     let currentVisibility = false;
-    
+
     // Check if health check is already in progress and capture current visibility
     setBackendLoading(prev => {
       if (prev.isChecking) {
@@ -133,7 +134,7 @@ function App() {
 
       if (response.ok) {
         console.log('✅ Backend is healthy - hiding loading modal');
-        
+
         setBackendLoading(prev => ({
           ...prev,
           isVisible: false,
@@ -148,7 +149,7 @@ function App() {
           console.log('🔄 Backend reconnected - refreshing data...');
           await refreshBackendData();
         }
-        
+
         return true;
       } else {
         throw new Error(`Health check failed with status: ${response.status}`);
@@ -187,15 +188,15 @@ function App() {
   useEffect(() => {
     const performInitialHealthCheck = async () => {
       // Show loading modal immediately for initial startup
-      setBackendLoading(prev => ({ 
-        ...prev, 
-        isVisible: true, 
+      setBackendLoading(prev => ({
+        ...prev,
+        isVisible: true,
         isReconnection: false // This is initial startup, not reconnection
       }));
-      
+
       // Wait a moment for the UI to update
       await new Promise(resolve => setTimeout(resolve, 500));
-      
+
       // Check backend health (will automatically refresh data if modal was visible)
       await checkBackendHealth();
     };
@@ -208,17 +209,17 @@ function App() {
     const interval = setInterval(() => {
       setBackendLoading(prev => {
         const timeSinceLastCheck = Date.now() - (prev.lastCheckTime || 0);
-        
+
         // Check more frequently when modal is visible, less frequently when not
-        const shouldCheck = prev.isVisible 
+        const shouldCheck = prev.isVisible
           ? !prev.isChecking // Every 5 seconds when modal is visible
           : timeSinceLastCheck > 30000 && !prev.isChecking; // Every 30 seconds when modal is hidden
-        
+
         if (shouldCheck) {
           console.log('🔄 Periodic health check triggered. Modal visible:', prev.isVisible);
           checkBackendHealth();
         }
-        
+
         return prev; // Don't actually update state, just check conditions
       });
     }, 5000); // Check every 5 seconds
@@ -230,7 +231,7 @@ function App() {
   useEffect(() => {
     const handleWindowFocus = async () => {
       console.log('🔍 Window focused - checking backend health silently...');
-      
+
       // Check backend health silently - only show modal if it actually fails
       const healthCheckResult = await checkBackendHealth();
       // Loading modal will be shown automatically by checkBackendHealth if it fails
@@ -239,7 +240,7 @@ function App() {
     const handleVisibilityChange = async () => {
       if (!document.hidden) {
         console.log('🔍 Document became visible - checking backend health silently...');
-        
+
         // Check backend health silently - only show modal if it actually fails
         const healthCheckResult = await checkBackendHealth();
         // Loading modal will be shown automatically by checkBackendHealth if it fails
@@ -264,7 +265,7 @@ function App() {
   const handleApiKeySubmit = async () => {
     // Refresh API key status after submission
     await checkApiKeys();
-    
+
     // If there's a pending model change, retry it now
     if (pendingModelChange.retryFunction) {
       console.log(`Retrying ${pendingModelChange.type} model change to '${pendingModelChange.model}' after API key update`);
@@ -285,7 +286,7 @@ function App() {
   useEffect(() => {
     // Listen for menu events from Electron
     const cleanupFunctions = [];
-    
+
     if (window.electronAPI) {
       const cleanupNewChat = window.electronAPI.onMenuNewChat(() => {
         setActiveTab('chat');
@@ -309,7 +310,7 @@ function App() {
       // Handle Electron window events - check backend health silently
       const cleanupWindowShow = window.electronAPI.onWindowShow(async () => {
         console.log('🔍 Electron window shown - checking backend health silently...');
-        
+
         // Check backend health silently - only show modal if it actually fails
         const healthCheckResult = await checkBackendHealth();
         // Loading modal will be shown automatically by checkBackendHealth if it fails
@@ -318,7 +319,7 @@ function App() {
 
       const cleanupAppActivate = window.electronAPI.onAppActivate(async () => {
         console.log('🔍 Electron app activated - checking backend health silently...');
-        
+
         // Check backend health silently - only show modal if it actually fails
         const healthCheckResult = await checkBackendHealth();
         // Loading modal will be shown automatically by checkBackendHealth if it fails
@@ -344,32 +345,32 @@ function App() {
     <div className="App">
       <div className="app-header">
         <div className="app-title">
-          <Logo 
-            size="small" 
-            showText={false} 
+          <Logo
+            size="small"
+            showText={false}
           />
           <span className="version">v0.1.5</span>
         </div>
         <div className="tabs">
-          <button 
+          <button
             className={`tab ${activeTab === 'chat' ? 'active' : ''}`}
             onClick={() => setActiveTab('chat')}
           >
             {t('tabs.chat')}
           </button>
-          <button 
+          <button
             className={`tab ${activeTab === 'screenshots' ? 'active' : ''}`}
             onClick={() => setActiveTab('screenshots')}
           >
             {t('tabs.screenshots')}
           </button>
-          <button 
+          <button
             className={`tab ${activeTab === 'memory' ? 'active' : ''}`}
             onClick={() => setActiveTab('memory')}
           >
             {t('tabs.memory')}
           </button>
-          <button 
+          <button
             className={`tab ${activeTab === 'settings' ? 'active' : ''}`}
             onClick={() => setActiveTab('settings')}
           >
@@ -380,7 +381,7 @@ function App() {
 
       <div className="app-content">
         {/* Keep ChatWindow always mounted to maintain streaming state */}
-        <div style={{ 
+        <div style={{
           display: activeTab === 'chat' ? 'flex' : 'none',
           flexDirection: 'column',
           height: '100%'
@@ -401,8 +402,8 @@ function App() {
         </div>
         {/* Keep ScreenshotMonitor always mounted to maintain monitoring state */}
         <div style={{ display: activeTab === 'screenshots' ? 'block' : 'none' }}>
-          <ScreenshotMonitor 
-            settings={settings} 
+          <ScreenshotMonitor
+            settings={settings}
             onMonitoringStatusChange={setIsScreenMonitoring}
           />
         </div>

frontend/src/components/SettingsPanel.js

@@ -984,7 +984,14 @@ const SettingsPanel = ({ settings, onSettingsChange, onApiKeyCheck, onApiKeyRequ
     'gemini-2.5-flash',
     'gemini-2.5-flash-lite',
     'gemini-1.5-pro',
-    'gemini-2.0-flash-lite'
+    'gemini-2.0-flash-lite',
+    'llama3.2',
+    'mistral',
+    'gemma2',
+    'qwen2.5-coder-32b',
+    'deepseek-r1-distill-llama-70b',
+    'deepseek-v3.1:671b-cloud',
+    'qwen3-vl:235b-cloud'
   ];
 
   // Combine base models with custom models
@@ -999,6 +1006,13 @@ const SettingsPanel = ({ settings, onSettingsChange, onApiKeyCheck, onApiKeyRequ
     'gpt-4o',
     'gpt-4.1-mini',
     'gpt-4.1',
+    'llama3.2',
+    'mistral',
+    'gemma2',
+    'qwen2.5-coder-32b',
+    'deepseek-r1-distill-llama-70b',
+    'deepseek-v3.1:671b-cloud',
+    'qwen3-vl:235b-cloud'
   ];
 
   // Combine base memory models with custom models

mirix/agent/agent.py

@@ -725,6 +725,7 @@ def _get_ai_reply(
                         f"Attempt {attempt} failed: {he}. Retrying in {delay} seconds..."
                     )
                     time.sleep(delay)
+                    continue
 
             except Exception as e:
                 log_telemetry(

mirix/agent/agent_wrapper.py

@@ -20,6 +20,7 @@
     GEMINI_MODELS,
     MAXIMUM_NUM_IMAGES_IN_CLOUD,
     OPENAI_MODELS,
+    OLLAMA_MODELS,
     TEMPORARY_MESSAGE_LIMIT,
 )
 
@@ -882,6 +883,8 @@ def _determine_model_provider(
             return "anthropic"
         elif model_name in OPENAI_MODELS:
             return "openai"
+        elif model_name in OLLAMA_MODELS:
+            return "ollama"
         else:
             raise ValueError(f"Invalid model provider: {model_name}")
 
@@ -962,6 +965,15 @@ def _create_llm_config_for_provider(
                     model_wrapper=None,
                     context_window=128000,
                 )
+
+        elif provider == "ollama":
+            llm_config = LLMConfig(
+                model=model_name,
+                model_endpoint_type="ollama",
+                model_endpoint=self.agent_config.get("ollama_base_url") or model_settings.ollama_base_url or "http://localhost:11434/v1",
+                model_wrapper=None,
+                context_window=128000,
+            )
         else:
             # Custom provider - use custom config or fallback to OpenAI-compatible
             if custom_agent_config is not None:
@@ -1051,7 +1063,7 @@ def set_memory_model(self, new_model, custom_agent_config: dict = None):
         """Set the model specifically for memory management operations"""
 
         # Define allowed memory models
-        ALLOWED_MEMORY_MODELS = GEMINI_MODELS + OPENAI_MODELS
+        ALLOWED_MEMORY_MODELS = GEMINI_MODELS + OPENAI_MODELS + OLLAMA_MODELS
 
         # Determine the effective provider
         provider = self._determine_model_provider(new_model, custom_agent_config)
@@ -1105,6 +1117,8 @@ def set_memory_model(self, new_model, custom_agent_config: dict = None):
             required_keys = ["GEMINI_API_KEY"]
         elif new_model in OPENAI_MODELS:
             required_keys = ["OPENAI_API_KEY"]
+        elif new_model in OLLAMA_MODELS:
+             required_keys = [] # Ollama usually doesn't require an API key
 
         return {
             "success": True,

mirix/agent/app_constants.py

@@ -18,6 +18,16 @@
     "gpt-5",
 ]
 
+OLLAMA_MODELS = [
+    "llama3.2",
+    "mistral",
+    "gemma2",
+    "qwen2.5-coder-32b",
+    "deepseek-r1-distill-llama-70b",
+    "deepseek-v3.1:671b-cloud",
+    "qwen3-vl:235b-cloud",
+]
+
 STUCK_TIMEOUT = 10
 RUNNING_TIMEOUT = 30
 TOTAL_TIMEOUT = 60

mirix/configs/mirix.yaml

@@ -1,2 +1,8 @@
 agent_name: mirix
 model_name: gemini-2.5-flash-lite
+# Ollama support - uncomment to use Ollama instead
+# model_name: qwen3-vl:235b-cloud
+# model_endpoint_type: ollama
+# model_endpoint: http://localhost:11434/v1
+# context_window: 16384
+# put_inner_thoughts_in_kwargs: true

mirix/configs/mirix_monitor.yaml

@@ -1,3 +1,9 @@
 agent_name: mirix
 model_name: gemini-2.0-flash
+# Ollama support - uncomment to use Ollama instead
+# model_name: qwen3-vl:235b-cloud
+# model_endpoint_type: ollama
+# model_endpoint: http://localhost:11434/v1
+# context_window: 16384
+# put_inner_thoughts_in_kwargs: true
 is_screen_monitor: true

mirix/configs/mirix_ollama.yaml

@@ -0,0 +1,6 @@
+agent_name: mirix
+model_name: qwen3-vl:235b-cloud
+model_endpoint_type: ollama
+model_endpoint: http://localhost:11434/v1
+context_window: 16384
+put_inner_thoughts_in_kwargs: true