ClawDesk - Agent2OS

The private Agent2OS for real work. Chat with any AI model, connect your favorite messaging apps, and run agents across your desktop, terminal, cloud VM, Raspberry Pi, or anywhere Rust runs.

Quick Start · Architecture · tmux Desktop · Security · Crates · Development · Contributing

⚠️ Active Development — Things Will Break

ClawDesk is under active development and is not yet stable. APIs, CLI commands, and data formats may change without notice. Expect rough edges, missing features, and occasional breakage. We're currently focused on testing and hardening the core runtime.

Contributors and testers are welcome! If you hit a bug, please open an issue. If you want to help, check the open issues or submit a PR.

ClawDesk - Agent2OS is a personal AI workspace you can run on your own devices. It is built for people who want one place to chat with AI, organize work, automate repetitive tasks, and keep useful assistants available across desktop, terminal, and server environments.

Instead of switching between browser tabs, hosted tools, scripts, and disconnected bots, ClawDesk gives you a single home for AI-powered work. You can use it as a desktop app, keep it running in the background, connect it to the tools and channels you already use, or run it headless on a remote machine.

With ClawDesk, you can:

• Talk to AI in a way that fits your workflow from desktop, terminal, or command line
• Automate busywork such as repeated prompts, routines, checks, and operational tasks
• Connect your messaging and work tools so agents can meet you where work already happens
• Keep more control over privacy and setup by running it on your own machine or server
• Use one system across laptop, cloud VM, Raspberry Pi, and Docker instead of learning separate tools for each place
• Run always-on assistants for background jobs, monitoring, and long-running workflows

Runs on macOS, Linux, Windows, cloud VMs (AWS/GCP/Azure/DO), Raspberry Pi, and any machine with Rust.

In short: ClawDesk is Agent2OS — a private place to run AI assistants for real work, on your computer or infrastructure, in a way that stays close to your tools, your data, and your workflow.

Inspired by OpenClaw, ClawDesk takes the same broad idea of a personal, always-available AI system and pushes it toward a native, local-first experience with fewer moving parts and stronger control over where it runs.

Project status: ClawDesk is in active development. Expect rapid changes, rough edges, incomplete documentation in some areas, and occasional regressions as major features land. If you hit bugs, missing docs, confusing behavior, or platform-specific issues, please report them. Contributors, testers, and feedback from real-world usage are actively wanted.

Simplify. Reduce friction. Ship one binary.

ClawDesk exists to make powerful AI tooling feel more like a real product and less like a pile of parts. The goal is simple: give one system that can be easy enough for everyday use, but flexible enough for automation, remote machines, and serious long-running work.

• Start from a desktop app when you want something simple
• Drop to the terminal when you want more control
• Run it in the background when you need always-on behavior
• Use the same core system across personal devices, servers, and edge hardware
• Avoid getting locked into a single model, a single interface, or a hosted service

• It stays close to how you already work. Desktop when you want visuals, terminal when you want speed, background services when you want automation.
• It keeps you in control. You choose where it runs, which models it uses, and how your data flows.
• It is built for more than chat. ClawDesk is meant for ongoing tasks, workflows, tools, channels, and assistants that keep working after one prompt.
• It scales from personal use to serious setups. Start on a laptop, then move the same system to a server, VM, Docker host, or Raspberry Pi.
• It does not trap you in one surface. Use the GUI, CLI, tmux workspace, TUI, gateway, or daemon depending on the job.

ClawDesk is designed to be useful to both everyday users and technical operators. You can ignore the deeper internals at first and simply think of it as one system for AI chat, automation, connected tools, and long-running assistants.

ClawDesk is not tied to a single screen or setup. You can use the same system in the way that makes sense for where you are working:

No display required. If you do not want a desktop app, ClawDesk still works well through the terminal, background services, remote servers, Raspberry Pi, Docker, and SSH sessions.

All persistent storage — agent state, sessions, audit logs, skill configs, and vector embeddings — is handled by SochDB, an embedded ACID-compliant database written in Rust. SochDB provides:

• Embedded & zero-config — no external database process, no connection strings.
• ACID transactions — crash-safe writes for audit chains and session state.
• Vector search — built-in cosine similarity for memory/recall without external vector DBs.
• Single-file storage — one portable database file per workspace.

Channels are messaging platform integrations that let your agent send and receive messages across different surfaces. ClawDesk implements 25+ channel adapters as Rust traits:

Each channel implements the Channel trait (clawdesk-channel crate), providing a uniform interface for message routing, group handling, allowlists, and DM pairing — regardless of the underlying platform.

Skills are modular capabilities that extend what an agent can do. ClawDesk's skill system is ported from OpenClaw's skill platform and reimplemented in Rust:

• 15 bundled skills — ship with the app (file ops, shell, browser, memory, cron, etc.).
• Hot-loading — activate/deactivate skills at runtime without restarting.
• Per-skill ACLs — each skill declares required permissions; the security scanner enforces them.
• Workspace skills — drop a SKILL.md into your workspace to add custom skills.
• Managed skills — install community skill packs from registries.

┌──────────────────────────────────────────────────────────────┐
│                    ClawDesk Runtime                          │
├──────────────┬───────────────────────────────────────────────┤
│  React UI    │              Tauri IPC Bridge                 │
│  (TypeScript)│  138+ commands · typed invoke() wrappers      │
├──────────────┴───────────────────────────────────────────────┤
│                      Rust Backend                            │
│  ┌─────────────┬──────────────┬──────────────┬────────────┐  │
│  │ Skill       │ Provider     │ Security     │ Agent      │  │
│  │ Registry    │ Registry     │ Scanner      │ Runtime    │  │
│  ├─────────────┼──────────────┼──────────────┼────────────┤  │
│  │ Tool        │ Audit        │ Identity     │ Tunnel     │  │
│  │ Registry    │ Logger       │ Contracts    │ (WireGuard)│  │
│  └─────────────┴──────────────┴──────────────┴────────────┘  │
│                      SochDB Storage                          │
└──────────────────────────────────────────────────────────────┘

The React frontend communicates with the Rust backend through 138+ typed IPC commands over the Tauri bridge. The backend manages agent lifecycle, security scanning, skill orchestration, and model routing — all in-process, no sidecar daemons.

# Clone the repository
git clone https://github.com/clawdesk/clawdesk.git
cd clawdesk

# Install UI dependencies
cd crates/ui && pnpm install && cd ../..

# Run in development mode
cargo tauri dev

# Build for production
cargo tauri build

Prebuilt CLI binaries are currently published for macOS and Linux. Windows CLI binaries are coming soon.

# Build the CLI
cargo build -p clawdesk-cli

# Interactive first-time setup
clawdesk init

# tmux workspace (multi-pane terminal experience)
clawdesk tmux setup              # Guided onboarding + auto-launch
clawdesk tmux launch             # 4-pane workspace layout
clawdesk tmux launch -l chat     # Focused chat layout
clawdesk tmux launch -l monitor  # Ops monitoring layout

# Start the gateway
clawdesk gateway run

# Chat with an agent
clawdesk agent run
clawdesk agent msg "hello"

# Run diagnostics
clawdesk doctor

cargo tauri dev and cargo tauri build automatically prepare the bundled llama-server sidecar used by Local Models. A machine-level llama-server install is not required.

If you need to skip the auto-download in offline CI, set CLAWDESK_SKIP_LLAMA_SERVER_DOWNLOAD=1 and provide the sidecar files under crates/clawdesk-tauri/binaries yourself.

ClawDesk includes a command-line interface for quick actions, automation, diagnostics, and terminal-first workflows. If you like scripting or want AI available outside the desktop app, the CLI is the fastest path.

clawdesk
├── gateway run              Start the HTTP gateway server
├── message send <text>      Send a message to the agent
├── channels status          Show channel connectivity
├── plugins {list, reload, info}
├── cron {list, create, trigger, delete}
├── config {set, get, backup, restore}
├── agent
│   ├── message <text>       Send a one-shot message
│   ├── run                  Interactive REPL session (Claude Code equivalent)
│   ├── add <id>             Add agent from TOML or wizard
│   ├── validate             Validate all agent definitions
│   ├── list                 List agents with routing table
│   ├── apply                Hot-reload agent definitions
│   └── export <id>          Export agent to TOML
├── skill
│   ├── list / info          Browse installed skills
│   ├── search <query>       Search the skill registry
│   ├── install <name>       Install a skill pack
│   ├── uninstall <name>     Remove a skill
│   ├── create <name>        Scaffold a new skill
│   ├── lint / test          Validate skill definitions
│   ├── audit / check        Security audit for skills
│   └── publish              Publish to registry
├── tmux
│   ├── setup                Guided onboarding + tmux launch
│   ├── launch               Launch tmux session (default: desktop — 10 screens)
│   ├── list                 List active ClawDesk sessions
│   ├── attach <session>     Attach to a session
│   ├── kill <session>       Kill a session
│   ├── layouts              Show available layout presets
│   └── keys                 Show tmux key bindings cheat sheet
├── tui                      Launch the ratatui terminal UI
├── login                    Authenticate with provider APIs
├── doctor                   Run diagnostics
├── init                     Interactive first-time setup wizard
├── completions <shell>      Generate shell completions
├── security audit           Run security audit (8 checks)
├── daemon
│   ├── run / install / uninstall
│   ├── start / stop / restart
│   ├── status               PID, uptime, health
│   └── logs                 Tail daemon logs
└── update
    ├── check                Check for newer version
    ├── apply                Download and install update
    └── rollback             Rollback to previous version

ClawDesk includes a built-in tmux workspace for people who want a richer terminal experience. It gives you a multi-window setup that mirrors the main parts of the desktop app, which makes it useful on remote machines, cloud VMs, and SSH sessions.

# First-time: guided onboarding → provider setup → layout selection → auto-launch
clawdesk tmux setup

# Quick launch the full desktop experience (10 screens)
clawdesk tmux launch

# Quick-start presets
clawdesk tmux launch --layout workspace    # 4-pane dev layout
clawdesk tmux launch --layout monitor      # 3-pane ops dashboard
clawdesk tmux launch --layout chat         # 2-pane focused chat

Navigate with Ctrl-B + 0..9, just like clicking sidebar items in the Tauri app:

clawdesk tmux list                 # List active sessions
clawdesk tmux attach clawdesk      # Re-attach to a detached session
clawdesk tmux kill clawdesk        # Clean up
clawdesk tmux keys                 # Show key bindings cheat sheet
clawdesk tmux layouts              # Show all layout options

Mouse support is enabled by default. Ctrl-B + z zooms any pane to full screen.

See the full tmux Desktop Guide for details.

The TUI gives you a focused terminal dashboard for monitoring, navigation, and interactive use without opening the full desktop app:

clawdesk tui                       # Launch with dark theme
clawdesk tui --theme light         # Light theme
clawdesk tui --theme high-contrast # High contrast

Screens: Dashboard, Chat, Sessions, Agents, Channels, Memory, Skills, Settings, Logs, Security

Controls: j/k scroll, i insert mode, Tab cycle screens, Ctrl+1-9 switch sessions, q quit

ClawDesk implements defense-in-depth. Every layer enforces its own boundaries.

ClawDesk tracks token costs in real-time with per-model pricing:

# Run all tests
cargo test --workspace

# Run tests for a specific crate
cargo test -p clawdesk-security
cargo test -p clawdesk-skills
cargo test -p clawdesk-agents

# Run with output
cargo test --workspace -- --nocapture

# Lint
cargo clippy --workspace

clawdesk/
├── Cargo.toml              # Workspace root
├── crates/
│   ├── clawdesk-tauri/     # Tauri app (commands.rs, state.rs, lib.rs)
│   ├── clawdesk-agents/    # Agent runtime
│   ├── clawdesk-providers/ # LLM providers
│   ├── clawdesk-skills/    # Skill system
│   ├── clawdesk-security/  # Security (scanner, audit, identity)
│   ├── clawdesk-tunnel/    # WireGuard tunnel
│   ├── clawdesk-gateway/   # HTTP gateway
│   ├── clawdesk-types/     # Shared types
│   ├── ui/                 # React frontend
│   └── ...                 # 18 more crates
├── README.md
└── LICENSE

The UI is located at crates/ui/ and built with:

• React 18 with hooks-based architecture
• TypeScript with strict mode
• Vite for development and building
• Tailwind CSS for styling

Key files:

• src/App.tsx — Main application component with 5 navigation views
• src/api.ts — Typed wrappers around Tauri invoke() calls
• src/types.ts — TypeScript interfaces matching Rust Serialize structs

ClawDesk is under active development, and help is welcome across the board:

• Bug reports — if something breaks, behaves oddly, or is unclear, please open an issue.
• Testing — especially on different operating systems, cloud VMs, Raspberry Pi, Docker, tmux, TUI, and headless deployments.
• Documentation improvements — missing setup steps, unclear explanations, platform notes, examples.
• Code contributions — features, fixes, refactors, tests, and performance work.

If you are using ClawDesk in real work, on unusual hardware, or in production-like environments, that feedback is particularly valuable.

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-feature)
3. Commit changes (git commit -am 'Add my feature')
4. Push to branch (git push origin feature/my-feature)
5. Open a Pull Request

Please ensure:

• cargo test --workspace passes
• cargo clippy --workspace has no warnings
• New code includes tests

This project is licensed under the MIT License — see the LICENSE file for details.

• OpenClaw — The original AI agent gateway (TypeScript) that inspired ClawDesk's architecture, skill system, and channel abstractions.
• SochDB — Embedded ACID vector database powering all of ClawDesk's persistent storage.
• Tauri — Desktop app framework that makes single-binary native apps possible.
• llmfit — LLM fine-tuning toolkit.
• llama-swap — Hot-swap proxy for llama.cpp model serving.
• llama.cpp — The inference engine behind ClawDesk's local model support.
• pi-mono — AI agent toolkit: coding agent CLI, unified LLM API, TUI & web UI libraries, Slack bot, vLLM pods.
• agency-agents — Multi-agent orchestration patterns.