LangChain-Basic

Basic tutorial to learn langchain

How to run (local)

$ uv run uvicorn app.main:app --reload

Run everything with Docker Compose

The default compose file runs everywhere (macOS Apple Silicon, Linux, Windows) without assuming a GPU. Build and start both the FastAPI backend (backend service) and the React frontend with one command:

docker compose up --build

• Frontend (Vite dev server): http://localhost:5173
• Backend is only reachable from inside the Docker network (the frontend container proxies to it).

The frontend container reads VITE_BACKEND_URL (defaults to http://backend:8000 in docker-compose.yml) so you can point it at a different API by setting an environment override:

VITE_BACKEND_URL=https://your-api-host docker compose up --build frontend

The backend purposely is not published to the host for safety. If you want to hit it directly (e.g., http://localhost:8000/docs), temporarily add ports: ["8000:8000"] under the backend service in docker-compose.yml or run docker compose run --service-ports backend. Stop the stack with docker compose down.

Environment variables

Populate the root .env file before running Docker Compose. The backend service imports everything from this file via env_file so secrets/config stay out of the image build.

# LLM
OLLAMA_BASE_URL="http://host.docker.internal:11434"

# Affine MCP
AFFINE_BASE_URL="http://host.docker.internal:3010"
AFFINE_API_TOKEN="your-api-token"
AFFINE_EMAIL="your-email@example.com"
AFFINE_PASSWORD="your-password"

# IDA MCP (SSE server running on host)
IDA_MCP_URL="http://host.docker.internal:8744/sse"

Note: host.docker.internal allows Docker containers to connect to services running on the host machine.

MCP Server Setup

The backend uses Model Context Protocol (MCP) to integrate with external tools.

Affine MCP

Affine MCP is automatically installed in the Docker image via npm install -g affine-mcp-server. It supports email/password authentication.

IDA MCP

Since IDA Pro is a GUI application with licensing requirements, it runs on the host machine. The Docker container connects to it via SSE transport.

On the host machine, start IDA Pro and run the MCP server:

uv run ida-pro-mcp --transport http://127.0.0.1:8744/sse

The Docker container will connect to host.docker.internal:8744 to communicate with IDA Pro.

GPU passthrough (Linux/NVIDIA hosts)

When you need the backend to offload work to an NVIDIA GPU, use the backend-gpu service that lives in the same compose file. Start it alongside the frontend like this:

docker compose --profile gpu up --build backend-gpu frontend

Only the services you request are created, so the default CPU-only backend stays stopped and port 8000 remains available. The GPU service inherits the same image/build but adds deploy.resources.reservations.devices plus the standard NVIDIA_VISIBLE_DEVICES / NVIDIA_DRIVER_CAPABILITIES env vars. Install the NVIDIA Container Toolkit on the host beforehand. On macOS (including Apple Silicon/M-series) Docker Desktop cannot expose GPUs, so keep using the default CPU service (docker compose up --build).