Implement LLM-based entity extractor

.dockerignore

@@ -0,0 +1,38 @@
+# Git
+.git
+.gitignore
+
+# Python cache/env
+__pycache__/
+*.py[codz]
+*.pyo
+*.pyd
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.venv/
+venv/
+env/
+
+# Local env/secrets
+.env
+.env.*
+!.env.example
+
+# Build artifacts
+build/
+dist/
+*.egg-info/
+
+# Runtime/local data
+uploads/
+
+# Frontend workspace is not needed for backend container runtime
+extension/node_modules/
+extension/.wxt/
+extension/.output/
+extension/.env
+
+# Editor/system files
+.DS_Store
+Thumbs.db

Dockerfile

@@ -0,0 +1,36 @@
+# syntax=docker/dockerfile:1.7
+
+FROM python:3.12-slim AS runtime
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    UV_COMPILE_BYTECODE=1 \
+    UV_LINK_MODE=copy
+
+WORKDIR /app
+
+# Runtime deps:
+# - ffmpeg is required by faster-whisper for voice transcription routes
+# - git is used by repo-related tooling
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    ffmpeg \
+    git \
+    ca-certificates \
+ && rm -rf /var/lib/apt/lists/*
+
+# Install uv binary for reproducible lockfile-based installs.
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /usr/local/bin/uv
+
+# Install Python dependencies first for better layer caching.
+COPY pyproject.toml uv.lock ./
+RUN uv sync --frozen --no-dev --no-install-project
+
+# Copy application code.
+COPY . .
+
+EXPOSE 5454
+
+HEALTHCHECK --interval=30s --timeout=5s --start-period=20s --retries=3 \
+  CMD python -c "import urllib.request; urllib.request.urlopen('http://127.0.0.1:5454/api/genai/health/', timeout=3)" || exit 1
+
+CMD ["uv", "run", "uvicorn", "api.main:app", "--host", "0.0.0.0", "--port", "5454"]

README.md

@@ -17,6 +17,67 @@ Unlike typical AI assistants, this agent:
 
 ---
 
+## Run With Docker
+
+This repository has two runtime parts:
+
+- Python backend (FastAPI + tools) -> Dockerized below.
+- Browser extension (WXT/React) -> runs on your host browser, not inside Docker.
+
+### 1) Prepare environment
+
+Copy the sample environment file and set your keys:
+
+```bash
+cp .env.example .env
+```
+
+Minimum required value for backend boot:
+
+- `GOOGLE_API_KEY`
+
+If you plan to use Google OAuth features (Gmail/Calendar login), also set:
+
+- `GOOGLE_CLIENT_SECRET`
+
+### 2) Build and run API container
+
+```bash
+docker compose up --build -d agentic-api
+```
+
+Health check:
+
+```bash
+curl http://localhost:5454/api/genai/health/
+```
+
+Stop container:
+
+```bash
+docker compose down
+```
+
+### 3) Optional: run MCP mode in Docker
+
+```bash
+docker compose --profile mcp up --build agentic-mcp
+```
+
+### 4) Run extension locally against Docker backend
+
+From `extension/`:
+
+```bash
+pnpm install
+echo "VITE_API_URL=http://localhost:5454" > .env
+pnpm dev
+```
+
+Then load the unpacked extension from `extension/.output/chrome-mv3`.
+
+---
+
 ## Architecture
 
 <img width="1973" height="1305" alt="Agentic Browser Architecture" src="https://github.com/user-attachments/assets/21dac6a5-c9d7-499a-8648-becdb4a04bba" />

core/llm.py

@@ -131,12 +131,18 @@ def __init__(
         elif "base_url_override" in config:
             final_base_url = config["base_url_override"]
         elif config.get("base_url_env"):
-            final_base_url = os.getenv(config["base_url_env"])
+            # Try env var first, then fall back to the pydantic-settings default
+            # (which already defaults ollama to http://localhost:11434)
+            final_base_url = os.getenv(config["base_url_env"]) or getattr(
+                get_settings(), config["base_url_env"].lower(), None
+            )
 
         if final_base_url:
             base_url_param_name = config["param_map"].get("base_url", "base_url")
             params[base_url_param_name] = final_base_url
         elif config.get("base_url_env") and not final_base_url:
+            # For providers that *need* a base URL, raise; but for ollama we
+            # have a sensible default in Settings so we should never reach here.
             raise ValueError(
                 f"Base URL for '{self.provider}' not found. "
                 f"Please provide it directly or set the '{config['base_url_env']}' environment variable."
@@ -196,6 +202,7 @@ def summarize_text(self, text: str) -> str:
 _default: LargeLanguageModel | None = None
 
 
+# Maps provider → secret name for API keys (cloud providers only)
 _PROVIDER_TO_SECRET = {
     "google": "google_api_key",
     "openai": "openai_api_key",
@@ -204,12 +211,18 @@ def summarize_text(self, text: str) -> str:
     "openrouter": "openrouter_api_key",
 }
 
+# Maps provider → secret name for base URLs (local/self-hosted providers)
+_PROVIDER_TO_BASE_URL_SECRET = {
+    "ollama": "ollama_base_url",
+}
+
 
 def _build_default(
     provider: str | None = None,
     model: str | None = None,
     temperature: float | None = None,
     api_key: str | None = None,
+    base_url: str | None = None,
 ) -> LargeLanguageModel:
     s = get_settings()
     p = (provider or s.default_llm_provider or "google").lower()
@@ -227,11 +240,21 @@ def _build_default(
             api_key = get_secrets_service().resolve_sync(secret_name) if secret_name else ""
         except Exception:
             api_key = getattr(get_settings(), secret_name, "") if secret_name else ""
+    # Resolve base URL for providers that need it (e.g. Ollama) — sync path.
+    if base_url is None:
+        base_url_secret = _PROVIDER_TO_BASE_URL_SECRET.get(p)
+        if base_url_secret:
+            try:
+                from services.secrets_service import get_secrets_service
+                base_url = get_secrets_service().resolve_sync(base_url_secret) or None
+            except Exception:
+                base_url = None
     return LargeLanguageModel(
         model_name=model or cfg.get("default_model"),
         api_key=api_key or "",
         provider=p,  # type: ignore[arg-type]
         temperature=temperature if temperature is not None else 0.4,
+        base_url=base_url,
     )
 
 
@@ -252,27 +275,42 @@ async def reload_default_llm() -> LargeLanguageModel:
     except Exception:
         override = None
     provider = (override or {}).get("provider")
+    p_lower = (provider or "google").lower()
+
+    # Resolve API key for cloud providers
     api_key: str | None = None
-    secret_name = _PROVIDER_TO_SECRET.get((provider or "google").lower())
+    secret_name = _PROVIDER_TO_SECRET.get(p_lower)
     if secret_name:
         try:
             from services.secrets_service import get_secrets_service
             api_key = await get_secrets_service().resolve(secret_name)
         except Exception:
             api_key = None
+
+    # Resolve base URL for self-hosted providers (e.g. Ollama)
+    base_url: str | None = None
+    base_url_secret = _PROVIDER_TO_BASE_URL_SECRET.get(p_lower)
+    if base_url_secret:
+        try:
+            from services.secrets_service import get_secrets_service
+            base_url = await get_secrets_service().resolve(base_url_secret) or None
+        except Exception:
+            base_url = None
+
     if override:
         try:
             _default = _build_default(
                 provider=provider,
                 model=override.get("model"),
                 temperature=override.get("temperature"),
                 api_key=api_key,
+                base_url=base_url,
             )
         except Exception as exc:
             print(f"Failed to apply LLM override {override!r}, falling back to env defaults: {exc}")
             _default = _build_default(api_key=api_key)
     else:
-        _default = _build_default(api_key=api_key)
+        _default = _build_default(api_key=api_key, base_url=base_url)
     return _default
 
 
@@ -289,7 +327,10 @@ def __getattr__(self, item):
         return getattr(self._target(), item)
 
     def __call__(self, *args, **kwargs):
-        return self._target()(*args, **kwargs)
+        # LangChain chat models (e.g. ChatOllama) no longer support the
+        # deprecated callable syntax `model(messages)`. Use .invoke() instead,
+        # which is the standard LangChain v0.2+ API for all providers.
+        return self._target().invoke(*args, **kwargs)
 
     def __repr__(self):
         try: