ClawWork

A workspace for OpenClaw — desktop and mobile.

Parallel tasks, structured artifacts, scheduled automation — and files you can actually find later.

Download · PWA · Quick start · Teams · What you get · What it stores · How it works · Repo layout · Roadmap · Contributing · Keynote

⚠️ Official Repository This is the official ClawWork project: https://github.com/clawwork-ai/clawwork

A copycat repository (ClawWorkAi/ClawWork) and website (clawworkai.store) have been found without proper attribution. Please be aware and use the official links above.

Official Website: https://clawwork-ai.github.io/ClawWork/

OpenClaw is powerful. Plain chat is a bad container for the kind of work it can do.

Once you have multiple sessions, long-running jobs, approval stops, generated files, recurring automation, and different gateways, chat history turns into mud. Status vanishes. Files vanish. Context vanishes.

ClawWork fixes that. Every task becomes a durable workspace with its own session, artifacts, controls, and history — laid out in a three-panel UI: tasks on the left, active work in the center, artifacts and context on the right.

One agent is useful. A coordinated team of agents is a workforce.

ClawWork Teams package multiple agents into a single deployable unit — roles, personalities, skills, and workflow included. A coordinator agent breaks down the task and delegates to worker agents, each running in its own sub-session. You see the full orchestration in real time.

skill → agent → team

teams/clawwork-dev/
├── TEAM.md                  # team metadata and workflow
└── agents/
    ├── manager/             # coordinator — orchestrates the team
    │   ├── IDENTITY.md      # role and prompt
    │   ├── SOUL.md          # personality and style
    │   └── skills.json      # skill dependencies
    ├── architect/            # worker — designs solutions
    ├── frontend-dev/         # worker — builds UI
    ├── core-dev/             # worker — builds core logic
    └── ...

• TeamsHub — browse and install community-contributed teams from a Git-native registry.
• Create manually — define agents, identities, and skills in a step-by-step wizard.
• AI Builder — describe what you need and let an LLM design the team structure, roles, and prompts for you.

Once installed, pick a team when creating a task. The coordinator takes over from there.

• Each task runs in its own OpenClaw session, so you can switch between parallel jobs without mixing context.
• Streaming replies, thinking traces, tool call cards, and artifacts live in one place instead of being buried in chat history.
• Gateway, agent, model, and thinking settings are scoped per task.
• Files produced by the agent are saved to a local workspace and stay easy to browse later.
• Risky exec actions can stop for approval before they run.
• Scheduled tasks run on their own cadence without you watching.

brew tap clawwork-ai/clawwork
brew install --cask clawwork

Prebuilt macOS, Windows, and Linux builds are available on the Releases page. The app updates itself — new versions download in the background and install when you quit.

No install required — open cpwa.pages.dev in any modern browser. Works on desktop and mobile, installable to your home screen. See #206 for details and feedback.

1. Start an OpenClaw Gateway.
2. Open ClawWork and add a gateway in Settings. Authenticate with a token, password, or pairing code. Default local endpoint: ws://127.0.0.1:18789.
3. Create a task, pick a gateway and agent, and describe the work.
4. Chat: send messages, attach images, @ files for context, or use / commands.
5. Follow the task as it runs, inspect tool activity, and keep the output files.
6. Use search, artifacts, and cron when the work stops being small.

• Parallel tasks with isolated OpenClaw sessions — archived tasks can be reopened later
• Per-gateway session catalogs
• Session controls that matter: stop, reset, compact, delete, and sync
• Background work that stays readable instead of collapsing into one long thread
• Set tasks on a schedule with cron, every, or at expressions — pick from presets or write your own, check run history, trigger manually anytime
• Export any session as Markdown to keep a clean record outside the app

• Streaming responses in real time
• Inline tool call cards while the agent works
• Progress and artifacts in a side panel
• See what you're spending — usage status per gateway, cost breakdown per session, 30-day rolling dashboard

• Multi-gateway support
• Per-task agent and model switching
• Manage your agents directly — create, edit, delete, and browse their workspace files without leaving the app
• Browse the full tools catalog for each gateway so you know what the agent can reach
• Thinking level controls and slash commands
• Approval prompts for sensitive exec actions
• Get notified when things happen in the background — task completions, approval requests, gateway disconnects — click a notification and it jumps back to the task. Per-event toggles so you control the noise.

• Context that is actually useful: images, @ file references, voice input, watched folders
• Point the app at up to 10 folders and it watches for changes, re-indexing automatically so your context stays fresh
• Local artifact storage
• Code blocks and remote images in assistant replies are auto-extracted and saved to your workspace — no manual copy-paste
• Full-text search across tasks, messages, and artifacts

• System tray support
• Quick-launch window with a global shortcut (Alt+Space by default, customizable)
• Keyboard shortcuts throughout the app
• The app updates itself in the background — you see progress in Settings and it installs on quit
• Zoom in or out to your comfort level, and the app remembers your preference
• Light and dark themes with 8 languages

• Export a debug bundle (logs, gateway status, sanitized config) when something goes wrong — useful for bug reports
• See which Gateway server version you're connected to right in Settings

Everything lives in a local workspace directory you choose. No cloud sync, no external database.

• Tasks — each one maps to an independent OpenClaw session, so parallel work never collides.
• Messages — user, assistant, and system messages, including tool calls and image attachments, all persisted locally.
• Artifacts — code blocks, images, and files the agent produces. Automatically extracted from assistant output so nothing gets lost.
• Full-text search — search across all of the above. Find that one code snippet from three weeks ago without remembering which task it came from.

ClawWork talks to OpenClaw through a single Gateway WebSocket connection. Each task gets its own session key, which keeps concurrent work isolated. Everything — tasks, messages, artifacts, search indexes — is stored locally in your workspace directory.

packages/shared/       — protocol types, constants (zero deps)
packages/core/         — shared business logic: stores, services, ports
packages/desktop/
  src/main/            — Electron main: gateway WS, IPC, DB, artifacts, OS integration
  src/preload/         — typed window.clawwork bridge
  src/renderer/        — React UI: components, layouts, stores, hooks, i18n
packages/pwa/          — Progressive Web App (browser + mobile)
docs/                  — design docs, architecture invariants
e2e/                   — Playwright E2E tests (smoke + gateway integration)
scripts/               — build and check scripts
website/               — project website (React + Vite)
keynote/               — presentation slides (Slidev)

Electron 34, React 19, TypeScript, Tailwind CSS v4, Zustand, SQLite (Drizzle ORM + better-sqlite3), Framer Motion.

• Voice input requires a local whisper.cpp binary and model.
• Auto-updater works on packaged builds only; dev mode skips updates.
• Context folder watcher: max 10 directories, 4 levels deep, files up to 10 MB.

✅ Already shipping:

• Multi-task parallel execution
• Multi-gateway with token, password, and pairing code auth
• Tool call cards and approval dialogs
• Slash commands and thinking controls
• File context, artifact browsing, and auto-extract
• Agent management and tools catalog
• Scheduled (cron) tasks
• Usage and cost dashboard
• Native desktop notifications
• Folder watcher with auto-reindex
• Session export and debug bundle export
• Auto-update
• Tray, quick launch, and voice input
• Gateway server version display
• PWA with offline support and mobile UI (cpwa.pages.dev)
• Linux packages (AppImage + deb)
• Teams: create, manage, and orchestrate multi-agent ensembles
• TeamsHub: Git-native marketplace for discovering and installing team definitions
• Skills: ClawHub-powered skill discovery and install
• AI Builder: LLM-assisted team creation

🔮 Next up:

• Conversation branching
• Artifact diff view
• Custom themes

The project is early and moving fast. If you want to help:

• Read DEVELOPMENT.md for setup and project structure
• Check Issues
• Open a Pull Request
• Run pnpm check before submitting — it gates lint, architecture, UI contract, renderer copy, i18n, dead code, formatting, typecheck, and tests.

Apache 2.0