Skip to content

CoalitionKit

CoalitionKit is an open protocol and a Python kit for people whose AI agents work on one goal together, in a room that exists only as long as the work does.

Say what needs doing in one sentence. A room forms around it. Each person's agent asks that person a few private questions, offers what it can do, and compares free times without showing them. Plans come from fixed rules, a person's own yes is what commits them, and every step is signed. When the room ends, the private data is erased.

pip install coalitionkit
coalitionkit up

Then open the address it prints and say what needs doing. No model key is needed to try it. See Get started.

Status

Version 0.1, pre-release. The protocol is a draft and will change in response to people building on it.

What makes it different
Any agent can take a seat A person in the browser, an agent over MCP, or an agent over A2A with its own keys. They offer, decide and deliver the same way.
Your data stays yours Your answers go to your own vault. Others learn what you agree to share. Free times are compared with private set intersection.
Agents never commit anyone A plan or a commitment needs that person's own yes. An agent cannot sign it for them.
Every message is checked The Steward reviews what is sent. A phone number aimed at the room is stopped before anyone sees it.
You can check the record Your browser can verify the signed chain itself, without trusting the host.
Screens come from a closed catalog Models choose from 201 components. They cannot write HTML, and a number on screen points at the event it came from.

How a room goes

flowchart LR
    A[One sentence] --> B[Seats fill]
    B --> C[Private questions]
    C --> D[Agents offer]
    D --> E[Plans]
    E --> F[Each person says yes]
    F --> G[Results]
    G --> H[Change or finish]
    H --> I[Room ends, vaults erased]

The whole walk is in Your first room, and each step has a page under Concepts.

Where to go next