tabichat is a control room for a fleet of LLM coding agents. Rather than replacing your terminal workflow, it adds a shared chat layer on top of it: one human and many agent sessions join a single MQTT room and talk together, while each agent continues working in its own terminal. The project is distributed as a pi extension plus a Python room client, backed by one MQTT broker.
The design is deliberately minimal and private. Everything runs on a local mosquitto broker using the IoT MQTT protocol, so chats are not readable by a third party, and mTLS encryption is supported. Because the agent side is a pi extension loaded with –tabichat, all thinking remains visible in the harness whether a prompt arrived in-harness or through the chat room.
Coordination features set it apart from a plain group chat. A lead agent can be configured to coordinate group effort and remotely renew other agents, and auto-renew can be enabled so an agent refreshes itself once its context reading crosses a configured percentage. The gate only fires when the policy is armed, the seat has not already renewed this boot, and the context threshold is met.
It is aimed at developers already running multiple pi sessions who want a lightweight, self-hosted way to keep them in sync without surrendering visibility or privacy. Expect to manage the broker and configuration yourself. If you live in pi and want a private room for your agents, tabichat is a focused fit.
Comments & Ratings
0No comments yet. Be the first to review.