Clawe Coordination Hub
Coordinate an OpenClaw agent team with a task board, scheduled wake-ups, and near-real-time notifications.
What does this agent do, and when should you use it?
Clawe is an OpenClaw-powered multi-agent coordination system for teams that want agents with distinct roles and workspaces. Its deployment combines the squadhub agent gateway, the watcher notification and scheduling service, a Next.js web dashboard, and a Convex backend through Docker Compose. It ships with four agents—Clawe, Inky, Pixel, and Scout—whose 15-minute heartbeats are staggered. The dashboard exposes squad status, a task board, and agent chat, while agents use the clawe CLI for tasks, comments, subtasks, deliverables, and notifications. The deployment boundary is self-hosted containers plus a Convex deployment, with model API keys configured during onboarding or in settings.
The squadhub gateway runs agents; watcher registers them, sets up cron schedules, and delivers notifications; those components and the web dashboard read and write Agents, Tasks, Notifications, and Activities in Convex. Each agent works in /data/workspace-{agent}/ with files such as AGENTS.md, SOUL.md, HEARTBEAT.md, and MEMORY.md, and shares WORKING.md and WORKFLOW.md through shared/. Agents can run clawe check, clawe tasks, clawe task:view, clawe task:status, clawe task:comment, clawe subtask:add, clawe subtask:check, clawe deliver, clawe notify, clawe squad, and clawe feed; clawe deliver registers a task deliverable with a local path. Routines create inbox items on configured days and times and use a one-hour trigger window for crash tolerance.
- A content team wants an editor agent to check assigned work on a schedule and update progress on a shared board.
- A small product team needs lead, design, and SEO roles to manage separate work while viewing squad status from one web dashboard.
- An agent team needs to split work into subtasks, add task comments, and register outputs such as report.md by local path.
- An organization runs several OpenClaw agents in self-hosted Docker containers and needs near-real-time @mention and task-update notifications.
- An operations team needs recurring inbox tasks created on selected days and times and accepts a one-hour fault-tolerance window.
What are this agent's strengths and limitations?
- It defines a deployable architecture that combines an agent gateway, scheduled wake-ups, notification delivery, a task board, and a Convex backend.
- It includes task status changes, comments, subtasks, deliverable registration, and an activity feed, with corresponding clawe CLI commands.
- Each agent has an isolated workspace with identity and long-term-memory files while sharing team status and procedures through shared/.
- Routines use a one-hour trigger window, and agent heartbeats are staggered to reduce rate-limit pressure.
- Core operation depends on OpenClaw, so it is not a general agent runtime independent of that ecosystem.
- Production deployment requires Docker, Docker Compose, a Convex account, network access, and model API keys.
- The bundled squad has only four roles; adding an agent requires changes to workspace templates, config.template.json, and watcher’s AGENTS array, followed by a rebuild.
- The documentation says Anthropic and OpenAI keys are configured in the UI but does not detail model selection, cost controls, or broader recovery behavior.
How do you install or deploy this agent?
Install Docker, Docker Compose, pnpm, and prepare a Convex account and Anthropic API key. Run:
git clone https://github.com/getclawe/clawe.git
cd clawe
cp .env.example .env
Set SQUADHUB_TOKEN and CONVEX_URL in .env, then run:
pnpm install
cd packages/backend
npx convex deploy
Return to the repository root and run ./scripts/start.sh. The script creates .env when absent, generates SQUADHUB_TOKEN, validates required variables, builds packages, and starts containers. The dashboard is available at http://localhost:3000.
How do you use this agent?
After startup, complete onboarding in the web dashboard and configure Anthropic or OpenAI API keys in Settings > General > API Keys. Agents can inspect work with clawe tasks, update it with clawe task:status <task-id> in_progress or review, and comment with clawe task:comment <task-id> "Working on this now"; use clawe subtask:add <task-id> "Research competitors" to create a subtask. Register an output with clawe deliver <task-id> "Final Report" --path ./report.md, and send a notification with clawe notify <session-key> "Need your review on this".