Mission control for your agent teams.
ClawKitchen is the interface layer over OpenClaw and ClawRecipes: a board you can actually watch, teams you can actually manage, and a record of everything your agents did.
openclaw plugins install @jiggai/kitchenPlugin installs
27,187 to date, per npm
Monthly installs
1,310 in August 2026, per npm
In production
Running, not a preview
What it looks like without this.
Every one of these is a process problem wearing a tooling costume. They are the reason the plugin exists.
I can’t tell what the agents are doing until something breaks.
A board with live status per card, so a stuck run is visible while it is still cheap.
Runs scroll past in a terminal and then they’re gone.
Every workflow run kept as a file, openable node by node, with its status and outputs still there tomorrow.
Adding an agent means someone editing a config file nobody else understands.
Roles and per-agent model and tool settings, editable without touching YAML.
Nobody can say what an agent actually changed last week.
Runs, tickets and outputs are files in the workspace, so last week is a directory you can open rather than a memory you argue about.
Every project rebuilds the same multi-agent setup from scratch.
Team templates you install version-pinned, instead of reassembling them each time.
An agent you cannot watch is an agent you cannot trust with anything that matters.
Supervision is not a nice-to-have on top of automation — it is the thing that decides how much you are willing to automate at all.
A UI for work that used to live in a terminal.
Agent runs are hard to supervise when the only interface is a log stream. ClawKitchen puts the whole operation on one screen.
The board
Every piece of work moves through swim lanes your team already understands — Backlog, In Progress, Testing, Done. Agents pick up cards; you watch them move.
- Lanes read straight from the files
- Per-card verification checklists
- Live status as agents work
Teams and agents
Assemble a line of specialists — Lead, Dev, DevOps, QA — and assign them to a workspace. Change who does what without touching a config file.
- Prebuilt specialist roles
- Per-agent model and tool config
- Workspace isolation per team
Runs and history
Every workflow run is recorded as a file: which nodes fired, what each one returned, where it stopped for approval. Open last Tuesday the way you would open a build log.
- Run detail, node by node
- Status, outputs and approval state
- Filter across teams and workflows
Marketplace
Browse and install team and agent templates that already work, instead of assembling a multi-agent setup from scratch every time.
- Install proven team templates
- Share templates across projects
- Version-pinned installs
What you are actually looking at.
One screen that answers the three questions you have about a running agent team.
What is happening right now
Cards in flight, which agent holds each one, and how long it has been there. The stuck ones stand out without anyone going looking.
What happened before
A run history you can filter and open node by node, so "what happened on Tuesday" is a lookup rather than an investigation.
Who is on the line
Your teams and their roles, with per-agent model and tool settings you can change between runs.
What to reuse
Templates for the setups that worked, version-pinned so installing one next month gets you the same thing.
Three commands, then it is running.
It is an OpenClaw plugin, so there is no separate service to stand up and nothing to host.
openclaw plugins install @jiggai/kitchen01✓ installed @jiggai/kitchen
The plugin, registered with your existing OpenClaw install.
openclaw kitchen open02→ http://localhost:7777
The board, in your browser, reading the workspace you are already in.
openclaw recipes scaffold-team development-team -t devteam --apply-config03✓ ok · 5 agents · workspace-devteam
A real team on disk — lead, dev, devops, QA and a workflow runner. That command is ClawRecipes; Kitchen is where you then watch what it made.
Who this is for — and who it isn't.
It is a real tool with real edges. Better you find out here than three days in.
- You already run OpenClaw and have more than one agent working at a time
- More than one person needs to see what the agents are doing
- You care about being able to reconstruct what changed and why
- A single agent on a single machine — the terminal is genuinely fine
- You want a hosted service. This runs locally, beside your work
- You need approval workflows and RBAC today. Not there yet
ClawRecipes
ClawRecipes defines the teams, roles and workflows in markdown. ClawKitchen is where you watch them run. Most people end up with both.
The questions we get asked.
Does this replace OpenClaw?
No. It is a plugin for it. OpenClaw stays the thing that runs agents; ClawKitchen is the interface over the top. If you uninstall the plugin, your agents carry on exactly as before.
Where does it run, and what does it see?
Locally, on the machine that already has your workspace. It reads the same files and run history OpenClaw writes. Nothing is uploaded to us, and there is no account to create.
Do I need ClawRecipes as well?
No, but they were built to be used together. ClawRecipes defines the teams and workflows as markdown in a workspace directory; ClawKitchen watches them run. Either works on its own.
Is it stable enough for real work?
It is what we use to run our own agent teams every day, and the install count is npm’s rather than ours. That is the honest measure available — there is no enterprise support contract behind it.
What does it cost?
Nothing. The plugin is free and public. We make money consulting, and the plugins are the proof that we run this stuff ourselves.
Install it in one command.
ClawKitchen is an OpenClaw plugin. If you already run OpenClaw, you are one install away. ClawRecipes pairs with it.
openclaw plugins install @jiggai/kitchenWant this running on your own work?
The plugin is free and public. If you would rather someone set it up around your operation, that is what the consulting side does.