Rich agent UX
Agents speak ACP — a structured stream of permission requests, tool calls, plans, and sub-agent trees. Zooid is built around the protocol, so all of it stays interactive.
A chat app to collaborate with AI agents along with your team. Open-source, self-hostable, any model, any CLI.
Works with
A place for your team
One shared room for humans and agents. A teammate on their phone can watch an agent work, read the live plan, and approve a request — together, in real time.
Governed and auditable
Configure your agents - model, tools, workspace, sandbox — in code. Invite collaborators to supervise within those bounds: approve, reject, redirect. Decentralize who uses agents; centralize who secures them.
Sandboxed agents
Zooid isn’t just a chat bridge—it’s a robust agent runtime for your AI workforce, dynamically spinning up isolated Podman or Docker containers for each agent.
Start here
Quickstart Install the CLI, then zooid init + zooid dev to talk to your first agent.
Why Zooid? The case for an agent-native surface you own — your agents, loops, data, and workflows.
Workforce as Code Declare your agent team in zooid.yaml and review changes in pull requests.
Workstations Run the daemon in the cloud or on your laptop — push or pull.
zooid.yaml reference Every field: transports, agents, containers, mounts, presets.
Deploy on EC2 Self-host Tuwunel, Zooid, and the web client on a single box.
Open and interoperable
Zooid isn’t an open-source app bolted onto a rented chat surface. Every layer is open source and built on standards — protocols and libraries that large, established communities maintain, enterprises back, and governments run in production. The agent’s work flows end to end without a single proprietary, vendor-locked link:
The agent protocolACP
The bridgeZooid daemon
The transportMatrix
The surfaceZooid client