Multi-Agent Code Generator

Autonomous multi-agent code generation system built with LangGraph, FastAPI, and a secure E2B sandbox.

Architecture

flowchart LR
    A[User Prompt] --> B[Planner]
    B --> C[Coder]
    C --> D[Executor]
    D --> E[Reviewer]
    E -- coding --> C
    E -- done / failed --> F[Result]

Loading

What is included

• Planner, coder, reviewer, and executor agent loop
• LangGraph orchestration with retry cycles
• E2B-backed sandbox execution wrapper
• FastAPI SSE endpoint for live generation updates
• Next.js frontend with live timeline, file tree, Monaco editor, and sandbox output panel
• CI with pytest coverage gating
• Evaluation script that writes a markdown report for resume metrics

How to run

Backend

1. Create a virtual environment and install dependencies:

pip install -r requirements.txt

1. Copy the env template and set your keys:

copy .env.example .env

1. Start the API:

uvicorn api.main:app --reload --port 8000

1. Run the tests:

pytest

1. Generate the evaluation report:

python -m agents.evaluation

Frontend

1. Move into the frontend/ folder.
2. Install dependencies:

npm install

1. Create .env.local with NEXT_PUBLIC_API_BASE_URL=http://localhost:8000.
2. Start the UI:

npm run dev

Results

The evaluation script writes reports/eval_report.md with the actual metrics from the current model setup. Use those values directly for resume bullets and do not fabricate them here.