ClawKitchen Open Source
ClawKitchen is the local-first UI for operating a machine that runs OpenClaw and ClawRecipes. It turns terminal commands, workspace folders, config files, markdown docs, workflow runs, approvals, tickets, goals, and cron jobs into one understandable control surface for real client operations.
Why ClawKitchen
Most organizations start with individual assistants. That works for experiments, but client operations need structure: clear roles, durable memory, repeatable workflows, human checkpoints, logs, approvals, and artifacts. ClawKitchen keeps the source of truth file-backed and human-readable while making the day-to-day system visible.
Best fit
Use ClawKitchen when you are turning AI from demo into a managed service, internal team, or repeatable client workflow that needs to be reviewed, improved, and trusted.
Product pillars
Create OpenClaw teams with leads, specialists, QA, devops, researchers, marketers, support agents, writers, analysts, and custom roles. Each role gets its own workspace, instructions, tools, memory, and operating lane.
Use case: Example: launch a marketing team with strategist, copywriter, designer, publisher, and QA roles for a client content pipeline.
Keep project knowledge in readable files instead of invisible chat history. Teams can share memory, tickets, docs, brand rules, procedures, and outputs that survive restarts.
Use case: Example: store brand voice, offer details, social calendar rules, and campaign lessons where every workflow run can use them.
Design repeatable flows with visual nodes: start/end, LLM, tool, media-image/video/audio, human approval, writeback, and handoff. ClawKitchen edits file-first workflow definitions while ClawRecipes executes the runtime semantics.
Use case: Example: intake a blog topic → research → outline → draft → generate image → QA → publish handoff.
Turn client work into visible queues: backlog, in progress, testing, done. Agents can pick up tasks, attach artifacts, and keep the delivery process auditable.
Use case: Example: a dev team receives a feature request, breaks it into tickets, implements it, tests it, and waits for approval before closing.
Use builtin and custom ClawRecipes to package repeatable team setups, tools, procedures, prompts, workflows, and cron jobs. Start from shipped defaults, then evolve client-specific recipes when the pattern becomes yours.
Use case: Example: install a sales-research recipe, then adapt its prompts, sources, CRM fields, and handoff format for a specific agency.
Run recurring checks, content loops, reporting jobs, lead monitors, and maintenance tasks on schedule. Agents can work while humans sleep, then surface what matters.
Use case: Example: every weekday morning, collect leads, draft outreach notes, summarize priority replies, and queue human review.
Manage messaging bindings that power notifications, approval flows, reminders, and human-in-the-loop decisions. Approval nodes can pause work until a human signs off or requests revision.
Use case: Example: send a Telegram approval request for a campaign draft, then resume the workflow only after the reviewer approves it.
Track higher-level goals that outlive a single ticket. Goals help teams keep long-running outcomes visible while individual tasks move through the board.
Use case: Example: keep a Q2 content growth goal visible while agents execute weekly research, drafting, publishing, and reporting tickets.
Inspect what happened in a workflow run: run status, approval state, node outputs, logs, generated deliverables, and handoff points. Runs answer the operator question: what happened?
Use case: Example: show a client exactly how a campaign post was researched, drafted, reviewed, and scheduled.
Operate against your own OpenClaw machine and workspaces. Keep sensitive client context, credentials, channel bindings, recipes, workflows, and memory under your infrastructure and review process.
Use case: Example: operate separate teams for multiple clients without mixing their files, tokens, or memory.
Operator surfaces
See installed agents, inferred team groupings, and the fastest path into inspecting a specific agent.
Scaffold agents or full teams from builtin and custom recipes, then evolve reusable client patterns.
Manage structure, role files, skills, cron, workflows, memory, tickets, goals, and operational visibility.
Create and edit visual, file-first workflow definitions tied to a team.
Inspect execution history, statuses, approvals, node outputs, and generated artifacts.
Move file-backed work through backlog, in-progress, testing, and done with comments and ownership.
Connect messaging surfaces for notifications, reminders, and approval flows.
Review scheduled automation installed through recipes and decide what should run automatically.
Control focused operational behavior such as scaffold and cron installation modes.
Workflow model
A ClawRecipes workflow run is a directory on disk with a run record, logs, node outputs, approvals, and deliverables. ClawKitchen gives that file-first runtime a visual editor and run inspector.
Assign prompts to specific team agents, use model overrides, and validate structured output fields.
Run actions such as file writes, outbound posts, messages, scripts, and external integrations.
Generate images, video, or audio through driver-backed providers discovered from ClawRecipes and installed skills.
Pause execution for review, send approval requests, capture declines, and route revisions back into the workflow.
Send work to another workflow or team, including specialized social publishing flows.
Pass data between nodes with variables such as {{run.id}}, {{workflow.name}}, and upstream node outputs.
Persist final outputs, notes, artifacts, or status updates back into the team workspace.
Debug failed steps, inspect outputs, and show clients how work moved through the system.
Client use cases
Plan content calendars, draft posts, generate media, review brand fit, and publish through approved channels with human checkpoints where they matter.
Give AI teammates a structured engineering process: ticket intake, implementation, tests, CI review, PR summaries, and QA signoff.
Turn recurring research into a repeatable pipeline with source collection, synthesis, contradiction checks, and executive-ready outputs.
Package a client-specific operating system: collect requirements, install a starter team, add context files, configure workflows, then train humans on review points.
Starter teams
Lead, dev, devops, and test roles for file-first tickets, implementation, verification, and PR-style delivery.
SEO, copywriter, ads, social, designer, analyst, video, and compliance roles for campaign operations.
Platform-specific distribution and social execution workflows for publishing and review.
Citations-first research pipelines for market intelligence, competitor monitoring, and source-backed reports.
Brief-to-draft-to-edit pipelines for articles, documentation, newsletters, and long-form content.
Triage, resolution, escalation, and knowledge-base workflows for repeatable support operations.
Onboarding workflow
This is the client-facing implementation flow ClawKitchen is built to support. Start narrow, make the work visible, then scale after the first workflow earns trust.
Start with the outcome: what should happen, who approves it, what tools are involved, what data is sensitive, and what success looks like.
Deliverable: Client process map + automation opportunity list
Pick a narrow, valuable workflow and scaffold from a builtin recipe when possible. Avoid one giant bot. Build a small team with clear responsibilities, role memory, tools, and review gates.
Deliverable: Team blueprint with roles, files, tools, and permissions
Add brand docs, SOPs, product notes, examples, credentials references, ticket rules, channel conventions, and human preferences into the right memory layers.
Deliverable: Shared context workspace the agents can actually use
Turn the process into a visible workflow with LLM nodes, tool nodes, media nodes, handoffs, writebacks, scripts, and human approvals. Every step should be inspectable.
Deliverable: Runnable workflow with logs, artifacts, and failure visibility
Execute real work with humans in the loop. Capture mistakes as memory updates, tighten prompts, and add tests or checks where needed.
Deliverable: Pilot report with fixes, examples, and go-live checklist
Add cron schedules, additional teams, custom recipes, channel bindings, integrations, reporting, and run-review habits once the first workflow is trusted and repeatable.
Deliverable: Production operating cadence + expansion roadmap
Quick install
Install the UI plugin, restart the gateway, then enable Kitchen with explicit host, port, and auth settings.
Configuration
"kitchen": {
"enabled": true,
"config": {
"dev": false,
"host": "<ipAddress>",
"port": 7777,
"authToken": "<your_password>",
"qaToken": "<optional_qa_token>"
}
}Screenshots
ClawKitchen is intentionally visual: teams, workflows, tickets, runs, and artifacts should be easy to explain to a client or teammate.
Team management
Workflow visibility
Ready to build
ClawKitchen gives you the structure to discover, configure, run, observe, and improve agent teams as a real service.
Newsletter
Monthly updates on new recipes, workflows, and agent operations best practices.