Local OpenAI-compatible (Ollama) settings tile

app/Dockerfile

@@ -15,7 +15,8 @@ ENV PYTHONDONTWRITEBYTECODE=1 \
     CBL_DB_DIR=/app/data \
     CBL_DB_NAME=cb_tools_db \
     PORT=8080 \
-    CBQA_LOG_FILE=/app/logs/cbqa.log
+    CBQA_LOG_FILE=/app/logs/cbqa.log \
+    CBQA_IN_DOCKER=1
 # Note: STORAGE_BACKEND is no longer needed — Couchbase Lite is the only
 # supported storage backend. The Couchbase Server SDK has been removed.
 # Note: SKIP_TOON_INSTALL is intentionally NOT set. `python-toon` is now

app/ai_analyzer.py

@@ -11,6 +11,7 @@
 - Automatic garbage collection
 """
 
+import os
 import time
 import hashlib
 import secrets
@@ -50,12 +51,52 @@
     OPENAI_SDK_AVAILABLE = False
     logger.warning("OpenAI SDK not installed. Please run: pip install openai")
 
+# Local OpenAI-compatible servers (Ollama, LM Studio, vLLM, llama.cpp).
+# These speak the OpenAI Chat Completions API and typically need no real key.
+LOCAL_OPENAI_COMPAT_IDS = frozenset({"local-openai", "ollama", "lmstudio", "vllm"})
+
+
+def is_local_openai_compat(provider: str) -> bool:
+    return (provider or "").lower() in LOCAL_OPENAI_COMPAT_IDS
+
+
+def running_in_docker() -> bool:
+    """True when Flask is inside a container (cannot use localhost for host services)."""
+    flag = os.environ.get("CBQA_IN_DOCKER", "").lower()
+    if flag in ("1", "true", "yes"):
+        return True
+    return os.path.exists("/.dockerenv")
+
+
+def rewrite_ai_url_for_runtime(url: str) -> str:
+    """Rewrite localhost/127.0.0.1 to host.docker.internal when running in Docker.
+
+    Ollama/LM Studio run on the laptop, not in the container. Docker Desktop
+    exposes the host as host.docker.internal; compose also sets extra_hosts.
+    """
+    if not url:
+        return url
+    if not running_in_docker():
+        return url
+    rewritten = url
+    for host in ("localhost", "127.0.0.1", "0.0.0.0"):
+        rewritten = rewritten.replace(f"://{host}", "://host.docker.internal")
+    if rewritten != url:
+        logger.info("rewrote AI URL for Docker host access")
+        logger.debug("ai_url_rewrite from=%s to=%s", url, rewritten)
+    return rewritten
+
+
+def dummy_key_for_local(api_key: Optional[str]) -> str:
+    """Ollama accepts any Bearer token; the proxy still sends Authorization."""
+    key = (api_key or "").strip()
+    return key if key else "ollama"
+
 # ============================================================================
 # PyInstaller Resource Path Helper
 # ============================================================================
 
 import sys
-import os
 
 def get_resource_path(filename: str) -> str:
     """
@@ -2318,6 +2359,12 @@ def call_ai_provider(provider: str,
     import json
 
     logger.info("calling_ai_provider provider=%s model=%s language=%s", provider, model, language)
+
+    api_url = rewrite_ai_url_for_runtime(api_url or "")
+    if is_local_openai_compat(provider):
+        api_key = dummy_key_for_local(api_key)
+        if not api_url:
+            api_url = "http://localhost:11434/v1"
 
     # Get dynamic max_tokens based on model
     max_tokens = get_max_output_tokens(provider, model)
@@ -2329,7 +2376,7 @@ def call_ai_provider(provider: str,
     # ---------------------------------------------------------
     # Option 1: Use OpenAI SDK (Preferred for OpenAI/Grok)
     # ---------------------------------------------------------
-    if (provider == 'openai' or provider == 'grok') and OPENAI_SDK_AVAILABLE:
+    if (provider == 'openai' or provider == 'grok') and OPENAI_SDK_AVAILABLE and not is_local_openai_compat(provider):
         try:
             logger.debug("using_openai_sdk provider=%s", provider)
             start_time = time.time()
@@ -2401,7 +2448,7 @@ def call_ai_provider(provider: str,
     # ---------------------------------------------------------
 
     # Format request payload for specific provider
-    if provider == 'openai' or provider == 'grok':
+    if provider == 'openai' or provider == 'grok' or is_local_openai_compat(provider):
         ai_request_payload = {
             'model': model,
             'messages': [
@@ -2416,8 +2463,10 @@ def call_ai_provider(provider: str,
             ]
         }
 
-        # OpenAI uses max_completion_tokens, Grok uses max_tokens
-        if provider == 'openai':
+        # OpenAI cloud uses max_completion_tokens + JSON mode.
+        # Grok and local OpenAI-compatible servers (Ollama, LM Studio) use max_tokens
+        # and often reject response_format=json_object.
+        if provider == 'openai' and not is_local_openai_compat(provider):
             ai_request_payload['max_completion_tokens'] = max_tokens
             ai_request_payload['response_format'] = {'type': 'json_object'}
         else:
@@ -2550,7 +2599,7 @@ def call_custom_ai_provider(
     logger.debug("custom_api_name %s", custom_config.get("name"))
     logger.debug("custom_api_url_configured")
 
-    url = custom_config.get('url', '')
+    url = rewrite_ai_url_for_runtime(custom_config.get('url', ''))
     model = custom_config.get('model', 'default')
     auth_type = custom_config.get('authType', 'none')
     request_template = custom_config.get('requestTemplate', '')

app/app_base.py

@@ -727,7 +727,7 @@ def analyze_with_ai():
             logger.debug("  Model: %s", model)
             logger.debug("  Has API Key: %s", bool(api_key))
 
-            if not api_key:
+            if not api_key and not ai_analyzer.is_local_openai_compat(provider):
                 logger.error("No API key configured for provider: %s", provider)
                 return jsonify({
                     'success': False,
@@ -1142,7 +1142,7 @@ def test_ai_api():
             )
         else:
             # Test built-in provider
-            if not api_key:
+            if not api_key and not ai_analyzer.is_local_openai_compat(provider):
                 return jsonify({
                     'success': False,
                     'error': 'API key is required'
@@ -1158,6 +1158,8 @@ def test_ai_api():
                     api_url = 'https://api.x.ai/v1'
                 elif provider in ('gemini', 'google'):
                     api_url = 'https://generativelanguage.googleapis.com/v1beta'
+                elif ai_analyzer.is_local_openai_compat(provider):
+                    api_url = 'http://localhost:11434/v1'
 
             # Set default models if not provided
             if not model:
@@ -1169,6 +1171,8 @@ def test_ai_api():
                     model = 'grok-3-mini'
                 elif provider in ('gemini', 'google'):
                     model = 'gemini-2.5-flash'
+                elif ai_analyzer.is_local_openai_compat(provider):
+                    model = 'llama3.2'
 
             # Determine endpoint
             endpoint = '/chat/completions'
@@ -1315,17 +1319,20 @@ def ai_api_call():
         max_retries = data.get('maxRetries', 3)
 
         # Validation
-        if not api_key:
+        if not api_key and not ai_analyzer.is_local_openai_compat(provider):
             return jsonify({
                 'success': False,
                 'error': 'API key is required'
             }), 400
+        if ai_analyzer.is_local_openai_compat(provider):
+            api_key = ai_analyzer.dummy_key_for_local(api_key)
 
         if not api_url:
             return jsonify({
                 'success': False,
                 'error': 'API URL is required'
             }), 400
+        api_url = ai_analyzer.rewrite_ai_url_for_runtime(api_url)
 
         # Build full URL — Gemini embeds {model}:generateContent in the path
         if provider in ('gemini', 'google') and model:

app/assets/css/main.css

@@ -762,6 +762,19 @@ body { display: block !important; height: auto !important; overflow: visible !im
   box-shadow: 0 0 0 2px rgba(0, 123, 255, 0.1);
 }
 
+.ai-provider-help {
+  font-size: 12px;
+  color: #666;
+  margin: 6px 0 0 0;
+  line-height: 1.45;
+}
+.ai-provider-help code {
+  background: #f0f0f0;
+  padding: 1px 5px;
+  border-radius: 3px;
+  font-size: 11px;
+}
+
 /* Default badge */
 .default-badge {
   background: #007bff;

app/assets/js/ai-client.js

@@ -16,7 +16,8 @@ export async function getDefaultAIProvider() {
         if (preferences && preferences.aiApis && Array.isArray(preferences.aiApis)) {
             const defaultProvider = preferences.aiApis[0]; // First is default
 
-            if (defaultProvider && defaultProvider.apiKey) {
+            const localCompat = defaultProvider && ['local-openai', 'ollama', 'lmstudio', 'vllm'].includes(defaultProvider.id);
+            if (defaultProvider && (defaultProvider.apiKey || localCompat)) {
                 Logger.info('Using default AI provider:', defaultProvider.name);
                 return defaultProvider;
             }
@@ -64,10 +65,12 @@ export async function callAI(options) {
     Logger.info('🤖 AI API Call:', { provider, model, endpoint });
 
     // Validation
-    if (!apiKey) {
+    const localCompat = ['local-openai', 'ollama', 'lmstudio', 'vllm'].includes(provider);
+    if (!apiKey && !localCompat) {
         Logger.error('API key is required');
         return { success: false, error: 'API key is required' };
     }
+    const resolvedKey = apiKey || (localCompat ? 'ollama' : apiKey);
 
     if (!apiUrl) {
         Logger.error('API URL is required');
@@ -83,7 +86,7 @@ export async function callAI(options) {
             body: JSON.stringify({
                 provider,
                 model,
-                apiKey,
+                apiKey: resolvedKey,
                 apiUrl,
                 endpoint,
                 method,
@@ -142,6 +145,10 @@ export async function callDefaultAI(userMessage, options = {}) {
         case 'deepseek':
         case 'openrouter':
         case 'together':
+        case 'local-openai':
+        case 'ollama':
+        case 'lmstudio':
+        case 'vllm':
             // OpenAI-compatible format
             endpoint = '/chat/completions';
             payload = {
@@ -204,7 +211,7 @@ export async function callDefaultAI(userMessage, options = {}) {
     return callAI({
         provider: provider.id,
         model: provider.model,
-        apiKey: provider.apiKey,
+        apiKey: provider.apiKey || (['local-openai', 'ollama', 'lmstudio', 'vllm'].includes(provider.id) ? 'ollama' : provider.apiKey),
         apiUrl: provider.apiUrl,
         endpoint,
         payload
@@ -233,6 +240,10 @@ export function extractAIResponse(result, provider) {
             case 'deepseek':
             case 'openrouter':
             case 'together':
+            case 'local-openai':
+            case 'ollama':
+            case 'lmstudio':
+            case 'vllm':
                 // OpenAI format: data.choices[0].message.content
                 return data.choices?.[0]?.message?.content || null;
 

app/assets/js/main-legacy.js

@@ -27636,7 +27636,8 @@ ${info.features.map((f) => `   • ${f}`).join("\n")}
 
                 // Add standard AI providers
                 aiApis.forEach((api, index) => {
-                    const hasApiKey = api.apiKey && api.apiKey.length > 0;
+                    const localCompat = ['local-openai', 'ollama', 'lmstudio', 'vllm'].includes(api.id);
+                    const hasApiKey = localCompat || (api.apiKey && api.apiKey.length > 0);
                     const status = hasApiKey ? '✅' : '❌';
                     const defaultTag = index === 0 ? ' (Default)' : '';
                     const modelInfo = api.model ? ` - ${api.model}` : '';