An open-source AI team operating system built on OpenClaw. Agents brainstorm ideas, debate approaches, score candidates, write specs, implement code, and review each other's work — just like a real engineering team.
Chief chat → Brainstorm meeting → Reviewer scoring → Implementation chain → Completed
clawhq_demo.mp4
A single AI agent has blind spots. A team doesn't.
When you ask one agent to "build a todo app," you get one perspective, one architecture, no second opinion. If there's a bad decision buried in the code, nobody catches it.
ClawHQ runs a team — and teams work differently:
| Single Agent | ClawHQ Team | |
|---|---|---|
| Ideas | One approach, no alternatives | 3 agents brainstorm, 5+ competing candidates |
| Quality | Self-review (marking your own homework) | Independent reviewers score with structured scorecards |
| Decisions | Implicit, no record | Full audit trail — who proposed what, who scored how |
| Failures | "It broke" | Failure timeline with event context and root cause |
| Visibility | Terminal logs | Real-time office view, dashboard, meeting history |
- 🖥️ Visual, not terminal — Watch your team work in a real-time isometric office, not scroll through logs
- 📊 Structured decisions, not black boxes — Scorecards, voting, and full audit trails for every choice
- 🏛️ Real meetings, not prompt chains — 13 meeting types with proposals, debate, and synthesis
- 🔥 Debug failures, not guess — Failure timeline with event replay and root cause context
- 📈 Observe everything — KPI metrics, time series charts, and alert system built in
- 🏢 Isometric Office View — PixiJS 2.5D office with agent sprites, desks, meeting room, and decorations
- 🧠 Chief Chat — Natural language control, proactive check-ins, inline actions, notifications
- 👥 Multi-Agent Teams — 6 roles: PM, Developer, Reviewer, Designer, DevOps, QA
- ⛓️ Chain Plans — Editable task pipeline (plan → implement → review) with visual chain editor
- 🏗️ Team Presets — Pre-configured team templates for common workflows
- 📋 13 Meeting Types — brainstorm, planning, review, retrospective, kickoff, architecture, design, sprint-planning, estimation, demo, postmortem, code-review, daily
- 🔬 Tech Spec Meeting — 4-panel discussion (CTO / Frontend Lead / Backend Lead / QA) with conflict detection and synthesis
- ⚖️ Decision System — Proposal comparison, reviewer scorecards, Devil's Advocate, approve/revise/reject
- 📦 Deliverables — 6 types (Web / Code / Report / Data / Document / API) with live preview
- ⏪ History Replay — Event timeline playback for full session audit
- 📊 Monitoring Dashboard — KPI metrics, time series charts, alert system
- 🔥 Failure Timeline — Debug operational issues with full event context
- 🏢 Meeting Lineage — Parent-child meeting relationships, candidate inheritance
- 🌐 i18n — Full English and Korean language support
git clone https://github.com/pizzalist/ClawHQ.git
cd ClawHQ/app
npm install
npm run devRequires OpenClaw for agent orchestration.
Open http://localhost:3000 and try Chief Chat:
Build me a todo app→ PM creates a specyes→ Dev implements, Reviewer evaluatesapprove→ Decision finalized, visible in dashboard
Frontend: React + PixiJS + Zustand
Backend: Express + WebSocket + SQLite
Runtime: OpenClaw sessions
UI ↔ WebSocket/API ↔ Task Queue ↔ Agent Runtime ↔ Results/Events
Tech Stack: React · PixiJS · Zustand · Express · WebSocket · SQLite · TypeScript
app/
├── packages/
│ ├── web/ # React + PixiJS UI
│ ├── server/ # Express + task orchestration + DB
│ └── shared/ # Shared types
├── scripts/ # Demo, fixture, healthcheck scripts
├── docs/ # Scenarios, release notes, checklists
├── README.md
└── package.json
PRs welcome! Good starting points:
- Fix bugs and add tests
- Improve scenario templates
- Add observability and quality gates
- Improve docs and onboarding
MIT