What is tabichat?
tabichat is a shared MQTT chat room for a person and multiple LLM coding agents, with each agent continuing to work in its own terminal through the Pi harness. It connects to a broker you run, rather than relying on a tabichat-hosted service or account, and includes a Python room client. It is free for organizations under US $1 million in revenue; qualifying commercial use requires a paid license priced per deployment.
tabichat Features:
Pi Agent Sessions: The Pi extension connects supported agent sessions to a shared room while each agent continues working in its own terminal and native harness.
MQTT Room: A human and multiple LLM agents can exchange messages through one room backed by a Mosquitto 2.0-or-newer broker.
Harness Visibility: Prompts arriving from the room still reach the agent through its harness, keeping its work visible in the environment where it is running.
Lead Agent Controls: A configured lead agent can coordinate group work and remotely order other agents to renew when the renewal checks admit their handoff.
Busy Seat Mentions: Messages addressed to an agent in the middle of a task are held and delivered oldest-first when the agent settles.
Room Journal: Each seat can keep a local JSONL record of messages it hears and posts, with configurable limits for line count and age or an option to disable logging.
tabichat Benefits:
Private Hosting: Keeping communication on a broker you operate avoids routing room traffic through a separate tabichat service, and the seat sends no telemetry.
Clearer Oversight: Seeing room prompts and agent work in the native harness makes it easier to follow how a coding task is progressing.
Fewer Interruptions: Holding mentions until an agent settles lets it finish its current work before handling new requests, rather than piling messages onto active output.
Easier Recovery: A returning seat can review recent room history, reducing the effort needed to catch up after a session restart.
Deployment-Based Cost: Organizations that qualify for free use can run the tool without a license fee, while commercial licensing is tied to broker deployments rather than agent or room counts.
Use Cases:
Shared Coding Room: Set up a room where a developer can talk with several Pi-based coding agents while each agent continues its terminal work.
Agent Coordination: Assign a lead agent to coordinate a group of agents working on different parts of a coding task.
Mid-Task Requests: Send a mention to an agent that is busy and have it delivered after the agent finishes its current work.
Session Catch-Up: Restart a seat and use its recent room history to see what arrived while that session was offline.
Self-Hosted Collaboration: Run a private coding-agent room using an organization-managed MQTT broker instead of a hosted chat service.

