Coasts — Finally, Each Agent Gets Its Own Localhost
If you have tried running multiple Claude Code or Cursor instances in parallel across git worktrees, you know the pain. Port 3000 is already taken. The database is shared. One agent changes break another runtime. It is a mess that wastes more time than the parallelism saves.
Coasts, a YC-backed startup that hit Show HN today, fixes this with a beautifully simple idea: give every worktree its own containerized runtime. Each coast gets its own ports, its own network, its own services. Built in Rust, MIT licensed, and it works with any agent that uses worktrees — Claude Code, Cursor, Codex, whatever ships next month.
The tech is Docker-in-Docker with about 200MB overhead per containerized host. You write a Coastfile at your project root pointing to your docker-compose, and Coasts handles the isolation. Switching between worktrees went from 2 minutes to 8 seconds according to users in the HN thread. It supports per-service hot reload strategies, shared services to reduce duplication, and built-in secrets management.
This matters because the future of coding agents is parallel execution. Claude Code already supports worktrees natively. Codex runs subagents. Every serious AI coding setup is heading toward multiple agents working on different branches simultaneously. But until now, the runtime layer was the weak link — agents could write code in isolation but could not test it in isolation.
Coasts is macOS-first right now, with Linux support coming. It is free and open source.
https://github.com/coast-guard/coasts
← Back to all articles
Coasts, a YC-backed startup that hit Show HN today, fixes this with a beautifully simple idea: give every worktree its own containerized runtime. Each coast gets its own ports, its own network, its own services. Built in Rust, MIT licensed, and it works with any agent that uses worktrees — Claude Code, Cursor, Codex, whatever ships next month.
The tech is Docker-in-Docker with about 200MB overhead per containerized host. You write a Coastfile at your project root pointing to your docker-compose, and Coasts handles the isolation. Switching between worktrees went from 2 minutes to 8 seconds according to users in the HN thread. It supports per-service hot reload strategies, shared services to reduce duplication, and built-in secrets management.
This matters because the future of coding agents is parallel execution. Claude Code already supports worktrees natively. Codex runs subagents. Every serious AI coding setup is heading toward multiple agents working on different branches simultaneously. But until now, the runtime layer was the weak link — agents could write code in isolation but could not test it in isolation.
Coasts is macOS-first right now, with Linux support coming. It is free and open source.
https://github.com/coast-guard/coasts
Comments