Launched this week
Repowire
Peer-to-peer mesh for AI coding agents
4 followers
Peer-to-peer mesh for AI coding agents
4 followers
Real projects span multiple repos, multiple runtimes, multiple machines. Today every agent operates in a sealed box: no shared context, no live channel, no way to ask the agent next door. Repowire is a live mesh for your agents. Local-first. Runtime-agnostic
When working across features spanning multiple repositories, I noticed myself often copy pasting the output from one agent into another agent ("What is the API contract like?", "What terraform do I need for a feature shaped like X?", etc.). I called this the human bottleneck.
So, I set out to remove the human-in-the-middle to see whether the output quality of the model improves and/or the speed. It did. Started off simple, added a tmux mcp to my claude code. As I started using it more, I created Repowire.
Today, Repowire connects your agents into a live mesh. Any agent can query, notify, or broadcast to any other. You manage the mesh from a dashboard, Telegram, or Slack. It's local-first, works across agent runtimes, and scales from 2 peers to 20+. Currently supports Claude Code, Codex, Gemini, Opencode. And because we support multiple kinds of agents, we end up observing quite interesting patterns, such as
Multi-repo coordination: Agents in different repos ask each other live questions like ask_peer("project-b", "what endpoints do you expose?") instead of relying on stale docs.
Cross-agent review: One peer builds, another reviews for code quality, security, and simplification. Works especially well across different agent runtimes.
Orchestrator: A dedicated coordinator peer dispatches tasks, tracks progress, and runs review cycles, making 10+ agent meshes manageable.
Worktree isolation: Spawn peers on git worktrees so each works on its own branch and PR in parallel, with no merge conflicts during development.
Mobile mesh management: A Telegram bot lets you dispatch work, check peer status, and message any peer from your phone.
Infrastructure-as-peer: A dedicated infra peer (k8s, DNS, cloud config) that project peers coordinate with directly for namespaces, deploys, and the like.
Overnight autonomy: Hand peers long-running tasks (migrations, refactors, test suites) and disconnect and they report back via Telegram or dashboard when you return.
---
I hope you enjoy the tool,
Prass