AI Agent

An agent built with Streamlit, LangChain, and MCP tools that can query github.

GitHub API Setup

To use the GitHub search functionality, you need to set up a GitHub API token:

1. Create a GitHub Personal Access Token:

   • Go to GitHub Settings > Developer settings > Personal access tokens
   • Click "Generate new token (classic)"
   • Select scopes: public_repo (for public repositories) or repo (for private repositories)
   • Copy the generated token

2. Configure the token:

   • Create a .env file in the project root
   • Add your token: GITHUB_TOKEN=your_actual_token_here
   • The .env file is already in .gitignore for security

Docker Setup

To run all components using Docker Compose:

# Build and start all services
docker-compose up --build

# Run in background
docker-compose up -d --build

# Stop all services
docker-compose down

Services

• Ollama (port 11434): LLM service (for mac run a local install as docker ollama is slow and doesn't use mac gpu)
• MCP Server (port 3001): Search service (an MCP tool for the AI agent)
• API Server (port 3002): FastAPI backend (to host the AI agent)
• Streamlit App (port 8501): Web interface (mostly vibe coded as throwaway prototype to test)

Access Points

• Web UI: http://localhost:8501
• API: http://localhost:3002
• Ollama: http://localhost:11434

First Run

You need to install ollama locally

# For Mac
brew install ollama

# For Linux
curl -fsSL https://ollama.com/install.sh | sh

# For Windows
# Download and install from: https://ollama.com/download/windows

You'll also need to pull a model in Ollama:

# qwen:4b is small enough and has reasoning capabilities
ollama pull qwen:4b