feat: add support for free models in environment-based API key management

mert-ergun · mert-ergun · commit 1d68e0d5cd4d · 2026-02-13T15:36:42.000+03:00
diff --git a/crossbar_llm/backend/config.py b/crossbar_llm/backend/config.py
@@ -51,6 +51,11 @@
     "openrouter": "OpenRouter",
 }
 
+PRODUCTION_FREE_ENV_MODELS = {
+    "gpt-5-mini",
+    "gemini-3-flash-preview",
+}
+
 
 def get_setting(key, default=None):
     """Get a setting from the configuration."""
@@ -73,13 +78,13 @@ def get_provider_for_model(model_name: str) -> str | None:
     """
     if not model_name:
         return None
-    
+
     from models_config import get_provider_for_model_name
-    
+
     display_name = get_provider_for_model_name(model_name)
     if not display_name:
         return None
-    
+
     display_to_provider = {
         "OpenAI": "openai",
         "Anthropic": "anthropic",
@@ -89,7 +94,7 @@ def get_provider_for_model(model_name: str) -> str | None:
         "OpenRouter": "openrouter",
         "Ollama": "ollama",
     }
-    
+
     return display_to_provider.get(display_name)
 
 
@@ -102,3 +107,12 @@ def get_api_keys_status() -> dict:
             value != "" and value != "default"
         )
     return status
+
+
+def is_env_model_allowed(model_name: str) -> bool:
+    """Return whether a model is allowed with api_key='env' in this environment."""
+    if not model_name:
+        return False
+    if IS_DEVELOPMENT:
+        return True
+    return model_name in PRODUCTION_FREE_ENV_MODELS
diff --git a/crossbar_llm/backend/main.py b/crossbar_llm/backend/main.py
@@ -24,6 +24,7 @@
     get_provider_env_var,
     get_provider_for_model,
     get_setting,
+    is_env_model_allowed,
 )
 from config import get_api_keys_status as get_api_keys_status_from_config
 from dotenv import load_dotenv
@@ -54,6 +55,28 @@
 from tools.conversation_store import get_conversation_store, ConversationTurn
 from models_config import get_all_models
 
+
+def get_free_models_for_environment() -> List[str]:
+    """
+    Return model names that can be used with server-managed (`api_key='env'`) keys.
+    Development: all models.
+    Production: only explicitly allowed free models with configured provider keys.
+    """
+    all_models = get_all_models()
+    if IS_DEVELOPMENT:
+        return [model for provider_models in all_models.values() for model in provider_models]
+
+    free_models: List[str] = []
+    for model_list in all_models.values():
+        for model_name in model_list:
+            if not is_env_model_allowed(model_name):
+                continue
+            provider_id = get_provider_for_model(model_name)
+            env_var = get_provider_env_var(provider_id or "")
+            if env_var and os.getenv(env_var):
+                free_models.append(model_name)
+    return free_models
+
 # Load environment variables
 load_dotenv()
 
@@ -747,6 +770,23 @@ async def generate_query(
     # Handle "env" API key by using the API key from .env
     api_key = generate_query_request.api_key
     if api_key == "env":
+        if not is_env_model_allowed(generate_query_request.llm_type):
+            Logger.warning(
+                "[API] /generate_query/ - Model not allowed for env API key",
+                extra={
+                    "request_id": query_log.request_id,
+                    "model": generate_query_request.llm_type,
+                    "environment": "production" if IS_PRODUCTION else "development",
+                },
+            )
+            finalize_query_log(status="failed")
+            raise HTTPException(
+                status_code=403,
+                detail=(
+                    f"Model '{generate_query_request.llm_type}' is not available "
+                    "with server-managed API keys in this environment."
+                ),
+            )
         if not provider:
             Logger.error(
                 "[API] /generate_query/ - Provider required for env API key",
@@ -1066,6 +1106,23 @@ async def run_query(
     # Handle "env" API key by using the API key from .env
     api_key = run_query_request.api_key
     if api_key == "env":
+        if not is_env_model_allowed(run_query_request.llm_type):
+            Logger.warning(
+                "[API] /run_query/ - Model not allowed for env API key",
+                extra={
+                    "request_id": query_log.request_id,
+                    "model": run_query_request.llm_type,
+                    "environment": "production" if IS_PRODUCTION else "development",
+                },
+            )
+            finalize_query_log(status="failed")
+            raise HTTPException(
+                status_code=403,
+                detail=(
+                    f"Model '{run_query_request.llm_type}' is not available "
+                    "with server-managed API keys in this environment."
+                ),
+            )
         if not provider:
             Logger.error(
                 "[API] /run_query/ - Provider required for env API key",
@@ -1421,6 +1478,19 @@ async def event_generator():
         # Handle "env" API key
         api_key = run_query_request.api_key
         if api_key == "env":
+            if not is_env_model_allowed(run_query_request.llm_type):
+                yield {
+                    "event": "failed",
+                    "data": json.dumps({
+                        "error": (
+                            f"Model '{run_query_request.llm_type}' is not available "
+                            "with server-managed API keys in this environment."
+                        ),
+                        "error_type": "ModelNotAllowed",
+                        "attempts": attempts
+                    })
+                }
+                return
             if not provider:
                 yield {
                     "event": "failed",
@@ -2078,6 +2148,17 @@ def get_available_models():
     return models
 
 
+@app.get("/free_models/")
+def get_free_models():
+    """
+    Get models that are available without user API key (server-managed env keys).
+    """
+    Logger.info("Free models requested")
+    free_models = get_free_models_for_environment()
+    Logger.debug(f"Returning {len(free_models)} free models")
+    return {"models": free_models}
+
+
 # Initialize logging on startup
 @app.on_event("startup")
 async def startup_event():
diff --git a/crossbar_llm/frontend/src/components/ChatLayout.js b/crossbar_llm/frontend/src/components/ChatLayout.js
@@ -59,7 +59,7 @@ import CloseIcon from '@mui/icons-material/Close';
 import SyntaxHighlighter from 'react-syntax-highlighter';
 import { docco, dracula } from 'react-syntax-highlighter/dist/esm/styles/hljs';
 import NodeVisualization from './NodeVisualization';
-import api, { getAvailableModels } from '../services/api';
+import api, { getAvailableModels, getFreeModels } from '../services/api';
 import axios from 'axios';
 import Fuse from 'fuse.js';
 import { loadSuggestions } from '../utils/loadSuggestions';
@@ -149,6 +149,7 @@ function ChatLayout({
   const [apiKeysStatus, setApiKeysStatus] = useState({});
   const [apiKeysLoaded, setApiKeysLoaded] = useState(false);
   const [modelChoices, setModelChoices] = useState({});
+  const [freeModels, setFreeModels] = useState([]);
   const [modelsLoaded, setModelsLoaded] = useState(false);
 
   // Expanded sections in right panel
@@ -247,10 +248,13 @@ function ChatLayout({
       try {
         const models = await getAvailableModels();
         setModelChoices(models);
+        const freeModelNames = await getFreeModels();
+        setFreeModels(freeModelNames);
         setModelsLoaded(true);
       } catch (error) {
-        console.error('Error fetching available models:', error);
+        console.error('Error fetching available/free models:', error);
         setModelChoices({});
+        setFreeModels([]);
         setModelsLoaded(true);
       }
     };
@@ -265,9 +269,6 @@ function ChatLayout({
         if (response.data) {
           setApiKeysStatus(response.data);
           setApiKeysLoaded(true);
-          if (provider && response.data[provider]) {
-            setApiKey('env');
-          }
         }
       } catch (error) {
         console.error('Error fetching API keys status:', error);
@@ -276,16 +277,17 @@ function ChatLayout({
     fetchApiKeysStatus();
   }, [provider, setApiKey]);
 
-  // When provider changes, update API key
+  // When provider/model changes, default to server key only for free models
   useEffect(() => {
     if (apiKeysLoaded && provider) {
-      if (apiKeysStatus[provider]) {
+      const selectedModelIsFree = !!llmType && freeModels.includes(llmType);
+      if (apiKeysStatus[provider] && selectedModelIsFree) {
         setApiKey('env');
-      } else {
+      } else if (apiKey === 'env') {
         setApiKey('');
       }
     }
-  }, [provider, apiKeysStatus, apiKeysLoaded, setApiKey]);
+  }, [provider, llmType, freeModels, apiKeysStatus, apiKeysLoaded, apiKey, setApiKey]);
 
   // Scroll to bottom when messages update
   useEffect(() => {
@@ -494,9 +496,11 @@ function ChatLayout({
   const isSettingsValid = useCallback(() => {
     if (!provider) return false;
     if (!llmType) return false;
-    if (!apiKeysStatus[provider] && !apiKey) return false;
+    const selectedModelIsFree = freeModels.includes(llmType);
+    const canUseServerKeyForSelection = apiKeysStatus[provider] && selectedModelIsFree;
+    if (!canUseServerKeyForSelection && !apiKey) return false;
     return true;
-  }, [provider, llmType, apiKeysStatus, apiKey]);
+  }, [provider, llmType, apiKeysStatus, freeModels, apiKey]);
 
   // Check if semantic search settings are valid (only when enabled)
   const isSemanticSearchValid = useCallback(() => {
@@ -550,7 +554,9 @@ function ChatLayout({
     abortControllerRef.current = new AbortController();
     const signal = abortControllerRef.current.signal;
 
-    const effectiveApiKey = (apiKeysStatus[provider] && apiKey === 'env') ? 'env' : apiKey;
+    const selectedModelIsFree = freeModels.includes(llmType);
+    const canUseServerKeyForSelection = apiKeysStatus[provider] && selectedModelIsFree;
+    const effectiveApiKey = (canUseServerKeyForSelection && (apiKey === 'env' || !apiKey)) ? 'env' : apiKey;
 
     try {
       // Build request data
@@ -636,7 +642,9 @@ function ChatLayout({
     abortControllerRef.current = new AbortController();
     const signal = abortControllerRef.current.signal;
 
-    const effectiveApiKey = (apiKeysStatus[provider] && apiKey === 'env') ? 'env' : apiKey;
+    const selectedModelIsFree = freeModels.includes(llmType);
+    const canUseServerKeyForSelection = apiKeysStatus[provider] && selectedModelIsFree;
+    const effectiveApiKey = (canUseServerKeyForSelection && (apiKey === 'env' || !apiKey)) ? 'env' : apiKey;
 
     try {
       const runResponse = await api.post('/run_query/', {
@@ -883,7 +891,9 @@ function ChatLayout({
     abortControllerRef.current = new AbortController();
     const signal = abortControllerRef.current.signal;
 
-    const effectiveApiKey = (apiKeysStatus[provider] && apiKey === 'env') ? 'env' : apiKey;
+    const selectedModelIsFree = freeModels.includes(llmType);
+    const canUseServerKeyForSelection = apiKeysStatus[provider] && selectedModelIsFree;
+    const effectiveApiKey = (canUseServerKeyForSelection && (apiKey === 'env' || !apiKey)) ? 'env' : apiKey;
 
     try {
       // Build request data
@@ -2230,6 +2240,7 @@ function ChatLayout({
                         return <MenuItem key={`label-${idx}`} disabled sx={{ opacity: 0.7, fontWeight: 'bold', fontSize: '0.85rem' }}>{m.label}</MenuItem>;
                       }
                       const isSupported = supportedModels.includes(m);
+                      const isFreeModel = freeModels.includes(m);
                       return (
                         <MenuItem
                           key={m}
@@ -2256,15 +2267,15 @@ function ChatLayout({
                               fontWeight: 700,
                             }}>★</Box>
                           )}
-                          {m}
+                          {m}{isFreeModel ? ' (Free)' : ''}
                         </MenuItem>
                       );
                     })}
                   </Select>
                 </FormControl>
 
                 {/* API Key Section */}
-                {apiKeysLoaded && apiKeysStatus[provider] ? (
+                {apiKeysLoaded && apiKeysStatus[provider] && freeModels.includes(llmType) ? (
                   <Paper
                     variant="outlined"
                     sx={{
diff --git a/crossbar_llm/frontend/src/services/api.js b/crossbar_llm/frontend/src/services/api.js
@@ -134,6 +134,16 @@ export const getAvailableModels = async () => {
   }
 };
 
+export const getFreeModels = async () => {
+  try {
+    const response = await instance.get('/free_models/');
+    return response.data?.models || [];
+  } catch (error) {
+    console.error('Error fetching free models:', error);
+    throw error;
+  }
+};
+
 // Export the refreshCsrfToken function for explicit usage
 export { refreshCsrfToken };
 
