Show HN: Locus – AI agents that ship your code
github.comHi HN,
I built Locus, an open-source project management platform designed for engineering teams that want to use AI coding agents.
The core idea: plan sprints and manage tasks in a cloud dashboard, but all code execution happens locally on your infrastructure. Your source code never leaves your machine.
How it works:
- locus plan: AI generates a sprint plan by analyzing your codebase. It uses multi-agent roles to break down directives into structured tasks with complexity estimates and risk assessments. You can reject and iterate on plans before approving.
- locus run: An agent picks up tasks from the sprint, executes them sequentially on a single branch, auto-commits after each task, and opens a PR when the sprint is done. Supports both Claude and Codex as backends.
- locus review: Review that understands your full codebase, not just diffs. Detects bugs, security issues, and performance problems. Posts review comments directly to GitHub PRs.
- locus discuss: Have architectural discussions with AI that has full project context. Insights and decisions are automatically extracted and saved.
- Telegram bot: Manage your agent remotely from your phone.
- VSCode extension: Chat with AI in your editor with full repo context.
The whole system is free, self-hostable, and works with your existing GitHub workflow. There's a one-command installation script for Ubuntu/Debian/macOS.
Tech stack: TypeScript monorepo with NestJS API, Next.js dashboard, and a Bun-bundled CLI.
I'm a solo developer building this because I was frustrated with the disconnect between planning tools and actual code execution.
Would love feedback on the approach and any features you'd want to see.
No comments yet.