Multi-Agent System using LangGraph

A production-ready, general-purpose multi-agent system using LangGraph for orchestration, Redis for state management, Celery for asynchronous tasks, and FastAPI for the backend.

Overview

This system allows collaborative AI agents to work together to solve complex tasks. Unlike simple chatbots, this framework orchestrates specialized agents—a Research Agent that gathers information and a Writing Agent that synthesizes it—to produce high-quality, comprehensive outputs for any user request.

Key Capabilities:

• Dynamic Research: Automatically identifies research topics from your prompt (e.g., "Compare X vs Y", "How to install Z").
• Adaptive Writing: intelligent template selection for Comparisons, Tutorials, Analyses, and Summaries.
• Human-in-the-Loop: Inspect and approve drafts before they are finalized.
• Production Architecture: Scalable, async design with persistent state and real-time updates.

Architecture

See ARCHITECTURE.md for detailed diagrams and component breakdown.

Quick Start

Prerequisites

• Docker & Docker Compose
• API Key for Groq or OpenAI

1. Configure Environment

Copy the example file and add your API key:

cp .env.example .env
# Edit .env:
# LLM_API_KEY=your_key_here

2. Start Services

Launch the entire stack (API, DB, Redis, Worker):

docker-compose up -d

3. Create a Task

The system handles natural language prompts automatically:

Comparison Task:

curl -X POST http://localhost:8000/api/v1/tasks \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Compare Redis vs PostgreSQL for caching"}'

Tutorial Task:

curl -X POST http://localhost:8000/api/v1/tasks \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Create a beginner tutorial for Docker setup"}'

Analysis Task:

curl -X POST http://localhost:8000/api/v1/tasks \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Analyze microservices vs monolithic architecture"}'

4. Monitor & Approve

Check task status and approve the draft:

# Get Status
curl http://localhost:8000/api/v1/tasks/{task_id}

# Approve
curl -X POST http://localhost:8000/api/v1/tasks/{task_id}/approve \
  -H "Content-Type: application/json" \
  -d '{"approved": true}'

Technology Stack

Component	Technology	Role
Orchestration	LangGraph	Manages agent state and workflow structure
API	FastAPI	Async REST endpoints & WebSockets
Queue	Celery	Asynchronous task processing
Database	PostgreSQL	Persistent storage for tasks and results
State/Cache	Redis	High-speed agent workspace & broker
LLM	LangChain	LLM abstraction (OpenAI / Groq)

Testing

Run the end-to-end test suite to verify system health:

docker-compose exec api pytest tests/test_e2e.py -v

License

MIT License.