Overview

Cuckoo is a chat app where AI agents are first-class, and an open protocol for connecting them. The hub moves messages between people and agent backends. It never runs a model: the intelligence, and its bill, belong to whoever owns the agent.

That means three things for you as a builder.

• You write an ordinary program. It connects out to the hub, receives messages, and replies. There is no inbound port to open and no queue to run.
• You keep your own state. The hub tells you a conversation id and a user id. Everything else about that person is yours to store.
• You can read every line of it. The hub is open source, so what it does with a message is something you can check rather than trust.

Where to start

If you want to…	Read
See something working in ten minutes	Quickstart
Understand the pieces before you type	Concepts
Attach your own harness to an agent in the app	Your own agent
Put an agent in front of customers	For companies
Know who is talking to your backend	Identity
Read the wire format	Protocol
See what the hub does with your data	Open source, one hub

What Cuckoo does not do

Being clear about this early saves you an afternoon.

• No inference. There is no model in the hub, no prompt, no key to add.
• No end-to-end encryption. Traffic is over TLS and messages are readable on the hub. It is a reliable relay, not a blind one.
• No identity beyond a name. Your backend learns a user id and a display name. Not an email, not a phone number, not their other chats.
• No agent-to-agent messaging yet. Agents talk to people. Teams of agents that talk to each other are planned, not built.
• No group chats or push notifications yet. Both are being built.

Status

Cuckoo is early. The protocol is versioned at /v1 and the parts documented here are built and tested, but they can still change while the version stays at zero. Anything not on this site does not exist yet, whatever a stray comment in the source may suggest.