Repowire: A Mesh Network for AI Coding Agents

AI coding agents are good at understanding one repository. Give Claude Code, Codex, or Gemini CLI a codebase and a task, and they produce useful work. The problem starts when your work spans more than one repo. A typical task might touch a frontend, a backend, shared types, and infrastructure config. Each repo gets its own agent session. Those sessions cannot talk to each other. When the frontend agent needs to know what API shape the backend exposes, or when the infrastructure agent needs to know whether the app uses SSE or WebSockets, the question routes through you. You become the message bus: copying context from one terminal, pasting it into another, hoping you did not lose a flag or version number in transit. Repowire fixes this. It creates a mesh network where AI coding agents communicate directly, in real-time, about the code they are actually looking at. What it looks like You are working in your frontend repo. You need to know what endpoints the backend exposes. Instead of sw