CodeForge-AI

CodeForge AI is a human-in-the-loop, multi-agent software engineering platform. It operates as a coordinated team of specialized AI agents that analyze software repositories, propose architectural implementation plans, generate patch diffs, run sandboxed tests, execute iterative feedback repair loops, and open Git branches and GitHub Pull Requests under strict human approval gates. The platform integrates with GitHub via a GitHub App, authenticating through installations to securely clone user repositories, map installations, and open Pull Requests without exposing secrets. It provides repository code intelligence through AST parsing for Python, TS/JS, and Go, dependency manifest extraction, semantic code chunking, and pgvector-backed hybrid semantic code search. The multi-agent workflow includes a Planner Agent for repository analysis and architectural plan creation, a Code Engineer Agent for generating structured file patches within strict size and safety boundaries, a Code Reviewer Agent for evaluating code quality and maintainability, a Security Reviewer Agent for scanning patches for secrets and vulnerabilities, a Test Engineer Agent for generating and running tests in an isolated sandbox, a Debug/Repair Agent for iterative bug-fixing, and an Orchestrator Agent that coordinates handoffs and enforces confidence thresholds. The system includes a durable job queue with real-time WebSocket progress streaming, a secure execution sandbox with path validation and secret-sanitized subprocess execution, and Git branch and Pull Request automation with protected branch enforcement and patch SHA-256 fingerprint verification. Production deployment features include structured JSON logging with secret redaction, security middlewares, health and readiness probes, and containerization with non-root multi-stage Docker containers, Gunicorn with Uvicorn ASGI workers, Nginx SPA proxying, and Docker Compose deployment stack.