The Idea
TL;DR
AI dev tools are powerful but integrating them with GitHub is painful. autosolve is a simple self-hosted bridge: poll → detect → dispatch → report.
The Problem
Tools like OpenHands, SWE-agent, Devin can write code. But connecting them to your GitHub workflow is surprisingly hard:
- Webhooks need infrastructure and break silently
- GitHub Actions cron is unreliable for long-running agents
- Most tools are designed for one-off runs, not continuous monitoring
- For small teams, the setup overhead kills the value
The Solution
A lightweight daemon that:
- Polls GitHub API on a schedule (no webhooks needed)
- Detects new issues, PRs, comments, labels matching your rules
- Dispatches any AI tool you want (ra-aid, Ollama, Python scripts, anything CLI-callable)
- Reports results back to GitHub — comments, PRs, status updates
All state is persisted locally. Missed nothing after restarts. Runs on a laptop or a Raspberry Pi.
Why Self-Hosted
- Your tokens stay on your machine
- No third-party services in the loop
- Works with private repos without exposing anything
- You control the AI tool, the rules, and the budget
Where We Are
The project is in active development. The core polling and scheduling infrastructure works. The AI dispatch layer is next.
Read the full original idea document for the complete vision, or check Status & Roadmap for what's done and what's planned.
Want to help shape this?
The architecture is still forming. Now is the best time to influence the direction. Open an issue or check the Ideas & Wishlist.