Add generic OpenAI-compatible provider for cheap/local model support

.env.example

@@ -27,5 +27,13 @@ DATABASE_URL=postgresql://smriti:smriti@localhost:5432/smriti
 # Default model for background intelligence (extraction, draft, review)
 OPENAI_MODEL=gpt-4o-mini
 
+# Generic OpenAI-compatible provider (Ollama, LM Studio, vLLM, Together, etc.)
+# Set the API URL to use a local or cheap model instead of OpenAI/Anthropic.
+# API key is optional — local servers like Ollama don't require one.
+# Then set background_intelligence.provider to "generic" in providers.yaml.
+# SMRITI_GENERIC_API_URL=http://localhost:11434/v1
+# SMRITI_GENERIC_MODEL=llama3.1:8b
+# SMRITI_GENERIC_API_KEY=
+
 # App
 DEBUG=false

README.md

@@ -184,6 +184,12 @@ cp .env.example .env
 # Leave keys commented out to use mock mode (no real LLM calls).
 # .env is the primary config path. config/providers.yaml is optional
 # and gitignored — you do not need it for a standard setup.
+#
+# Using Ollama or another local model? Set these instead:
+#   SMRITI_GENERIC_API_URL=http://localhost:11434/v1
+#   SMRITI_GENERIC_MODEL=llama3.1:8b
+# Then set background_intelligence.provider to "generic" in
+# config/providers.yaml. See .env.example for details.
 
 docker compose up -d postgres    # start the database
 make setup                       # backend venv + deps + migrations + CLI + frontend

backend/app/config_loader.py

@@ -65,6 +65,7 @@ class AppProviderConfig:
     openai: ProviderConfig = field(default_factory=ProviderConfig)
     anthropic: ProviderConfig = field(default_factory=ProviderConfig)
     openrouter: ProviderConfig = field(default_factory=ProviderConfig)
+    generic: ProviderConfig = field(default_factory=ProviderConfig)
     chat: ChatConfig = field(default_factory=ChatConfig)
     background: BackgroundConfig = field(default_factory=BackgroundConfig)
 
@@ -166,6 +167,29 @@ def load_config() -> AppProviderConfig:
         base_url_default="https://openrouter.ai/api/v1",
     )
 
+    # Generic OpenAI-compatible provider (Ollama, LM Studio, vLLM, Together,
+    # Groq, etc.). Enabled when base_url is set — api_key is optional because
+    # local servers like Ollama don't require one.
+    g = providers_raw.get("generic", {})
+    generic_base_url = (
+        os.environ.get("SMRITI_GENERIC_API_URL")
+        or g.get("base_url", "")
+    )
+    generic_api_key = (
+        os.environ.get("SMRITI_GENERIC_API_KEY")
+        or g.get("api_key", "")
+    )
+    generic_model = (
+        os.environ.get("SMRITI_GENERIC_MODEL")
+        or g.get("default_model", "")
+    )
+    generic = ProviderConfig(
+        enabled=bool(generic_base_url),
+        api_key=generic_api_key or "not-required",
+        default_model=generic_model,
+        base_url=generic_base_url,
+    )
+
     chat = ChatConfig(
         default_provider=os.environ.get("SMRITI_DEFAULT_PROVIDER")
             or chat_raw.get("default_provider", "openrouter"),
@@ -183,6 +207,7 @@ def load_config() -> AppProviderConfig:
         openai=openai,
         anthropic=anthropic,
         openrouter=openrouter,
+        generic=generic,
         chat=chat,
         background=background,
     )
@@ -248,7 +273,7 @@ def providers_status() -> dict[str, dict]:
             "status_label": "Ready" if getattr(cfg, name).enabled else "Disabled",
             "default_model": getattr(cfg, name).default_model,
         }
-        for name in ("openai", "anthropic", "openrouter")
+        for name in ("openai", "anthropic", "openrouter", "generic")
     }
     # Background intelligence uses one of the above providers. Report
     # its ACTUAL key/enabled/configured state instead of hardcoding True,

backend/app/providers/registry.py

@@ -108,6 +108,22 @@ def get_adapter(provider: str, allow_mock: bool = False) -> ProviderAdapter:
         except ImportError as e:
             raise ProviderNotConfiguredError(str(e))
 
+    if p == "generic":
+        # Generic OpenAI-compatible provider (Ollama, LM Studio, vLLM, etc.).
+        # Reuses OpenAIAdapter with the user-configured base_url. The api_key
+        # may be "not-required" for local servers — OpenAI SDK accepts any
+        # non-empty string.
+        if not cfg.base_url:
+            raise ProviderNotConfiguredError(
+                "Generic provider requires SMRITI_GENERIC_API_URL or "
+                "providers.yaml generic.base_url to be set."
+            )
+        try:
+            from app.providers.openai_adapter import OpenAIAdapter
+            return OpenAIAdapter(api_key=cfg.api_key, base_url=cfg.base_url)
+        except ImportError as e:
+            raise ProviderNotConfiguredError(str(e))
+
     raise ProviderNotConfiguredError(f"Unknown provider: {provider}")