Mesh network for AI coding agents
Stop the copy-paste dance. Enable Claude Code and OpenCode sessions to communicate directly across repositories in real-time.
Claude
I see you're working on the frontend. How can I help?
Repowire • Backend
@backend says:
Here are the current endpoints in `src/api.ts`:
- POST /auth/login
- GET /users/me
- GET /projects (paginated)
Features
A better way to manage context
Repowire bridges the gap between isolated AI sessions, creating a collaborative mesh of intelligence for your entire codebase.
Sync Communication
- Real-time messaging between active coding sessions. No stale context or manual updates.
Multi-Repo Context
- Ask questions about code in other repositories without leaving your current session.
Tmux Integration
- Seamlessly integrates with tmux to manage and discover active Claude sessions.
Daemon Architecture
- Central daemon manages peer discovery and routing, running as a robust system service.
Secure & Local
- All communication happens locally on your machine by default. No data leaves your network.
Auto-Discovery
- Sessions automatically find each other. Just start Claude in a repo and it joins the mesh.
How it works
Repowire sits between your Claude sessions, acting as a message broker.
Architecture
When you ask a peer a question, Repowire's daemon routes the query to the target session via tmux. It injects the query, waits for Claude to answer, and pipes the response back to you.
- 1
Query Injection
- You use the
ask_peertool. The daemon finds the target session in tmux and types the query for you. - 2
Context Retrieval
- The peer Claude instance reads its local files, understands the current state, and formulates an answer.
- 3
Response Delivery
- Hooks capture the response and send it back to your original session instantly.
Get Started in Seconds
Install with pip or uv. Requires macOS/Linux and Python 3.10+.
Alternatively: pip install repowire
See the documentation for full setup instructions.