Put a whole team of AI agents on it.
Marketing, research, ops, content, code. Give one brief to a team of agents and watch them split it up, hand work between each other, and finish it on a board you can see.
One request, split across specialists, tracked on the board.
A leader takes the request, delegates it, and narrates the work in chat.
- 01
Install and run
One npm install, then
clawboo. The dashboard opens on your machine. - 02
Add a key, or a CLI
Paste a provider key to use the built-in agents, or connect Claude Code, Codex or Hermes.
- 03
Give the team a task
A leader breaks it up, delegates, and every step lands on the board as it happens.
Bring the agents you already use. They finally talk to each other.
Each one is strong on its own and trapped in its own terminal, with its own memory and no record of what the others did. Clawboo gives all five one board, one chat and one memory, and any of them can lead.
- OpenClaw
- Hermes
- Claude Code
- Clawboo Native
- Codex
Nothing to install to start
Clawboo Native is built in and speaks to eleven providers. Bring a key from any one of them and a team is running before you install anything. Claude Code, Codex and Hermes take a key or a single sign-in; OpenClaw connects as a Gateway.
However they join, they coordinate through the board and structured messages, never by reading each other’s screen output, and each works in its own copy of the code so two agents can’t overwrite each other.
Read the runtimes guide
A durable board, with the chat running alongside it.
The board is where decisions land and the chat is the narration. Tasks survive restarts, two agents can never claim the same one, and a failed verification verdict blocks the move to done unless it is explicitly overridden.
One memory, shared
Facts and procedures every runtime on the team can read and write, instead of a private context window per runtime.
One board, every team
Work moves from backlog to done in columns you can see, scoped to one team or to all of them at once.
Agents you can open up
Model, prompt, tools and history for each teammate, editable in place.
See the whole team at once
Who delegated what to whom, live, across every team you are running.
Autonomous, but never unaccountable.
You can see what every agent did, what it cost, and why a task was allowed to finish.
- The builder is not the judge
- Work is checked by its own verify command, and delegated or large diffs get an independent reviewer on a detached, push-less copy of the repo. A missing verify command scores as a failed check, never as a silent pass.
- Spend is tracked, and you set the ceiling
- Budgets per agent, per team and per mission. Warn by default; a hard cap that auto-pauses is one setting away. Nothing is throttled behind your back.
- Limits live below the model
- Nesting depth, fan-out, lifetime children and create rate are enforced in code, not requested in a prompt. An agent cannot talk its way past them, and risky delegations wait for the leader.
- Local-first, with an address
- Your board and history are SQLite at ~/.clawboo/clawboo.db. Your keys are in an AES-256-GCM vault at ~/.clawboo/secrets/. No account, no product analytics.
- 5
- runtimes as peers
- 11
- providers built in
- 3
- operating systems, one CI matrix
Questions.
Is Clawboo free and open-source?
Yes, MIT licensed and free to use or fork. The code is on GitHub and the package is on npm.
Which agents can I use?
The built-in Native runtime speaks to eleven providers, and a key from any one of them is enough to start. Nothing else needs installing. Claude Code, Codex, Hermes and OpenClaw connect as optional peers.
Where does my data go?
Your board, chat and memory stay on your machine. Prompts and code go only to the model provider you configure. Clawboo itself has no account and no product analytics.
How do agents avoid stepping on each other?
Each task is claimed by a single conditional write, so two agents can never take the same one, and each works in its own copy of the repo.
What do I need to run it?
Node.js 22.12 or newer, on macOS, Linux or Windows. Connecting Hermes also needs Python 3.11 to 3.13 and pipx.
Is it ready for real work?
It is early and moves fast. The board, chat, runtimes, memory and verification gate all ship and are covered by end-to-end tests. Expect rough edges at the margins.