selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

GitHubWebsite
clawbox logo

Self-host clawbox

v4.2.0

Private AI assistant appliance on NVIDIA Jetson

ReleaseSource AvailableOther
GitHubWebsite
Install
Updated9 hours ago
LanguageTypescript
Setup difficultyModerate
MobileResponsive
Stars
32
CPU
NVIDIA Jetson Orin Nano 8GB (Super) / x86_64
RAM
8GB unified memory (Jetson)
Storage
512GB NVMe SSD
Ports
80 (Next.js web UI (production-server.js)), 18789 (OpenClaw Gateway (token-gated)), 3006 (Terminal WebSocket PTY server), 6080 (noVNC WebSocket (remote desktop)), 8090 (Hermes dashboard proxy), 3005 (ClawBox UI on x86_64 (configurable via CLAWBOX_PORT))
Install with Linux
ClawBox desktop with Chat, Settings, Files, Terminal, Coding Agent, ClawKeep, Memory Shard and Browser appsClawBox desktop — a Chrome OS-style environment served straight from the deviceClawBox chat panel answering on the desktop with provider, model and reasoning pillsCoding Agent run on the ClawBox desktop showing progress, changed files and pull requestSettings → Remote Control showing persistent clawbox.tech hostname and tunnel stateClawBox chat on a phone with thumb-sized microphone beside the text boxMemory Shard memory index card with file, chunk and source counts, health, and indexing scheduleSettings → Local AI showing one row per model on the boxSettings → Appearance Mascot Pet picker with built-in ClawBox crab and community galleryMemory Shard settings on Hermes edition showing embedding model with ClawBox AI cloud and on-box options

What is clawbox?

This personal AI assistant appliance from ID Robots Ltd. runs the OpenClaw or Hermes Agent on NVIDIA Jetson Orin Nano hardware. It provides a private, always-on AI server with 67 TOPS of AI performance, 512GB NVMe storage, and low power consumption. The operating system includes a 5-minute setup wizard via a WiFi hotspot, a desktop environment with windowed apps, and built-in applications for AI chat, a coding agent, file management, terminal, remote desktop, browser automation, and encrypted backups. It supports both local models (Gemma 4 E2B preinstalled via llama.cpp, plus Ollama) and cloud providers (Claude, GPT, Gemini, ClawBox AI) with per-conversation model switching. Multi-platform messaging integrates Telegram, Discord, WhatsApp, and email. Remote access uses a persistent Cloudflare tunnel with no Cloudflare account required. Backups encrypt snapshots to Cloudflare R2. The software uses a custom source-available license permitting personal non-commercial use; business use requires a separate agreement. Optional ClawBox AI Pro/Max subscriptions add the coding agent, memory indexing, cloud voice, and higher usage limits. An x86_64 installer allows running the same web OS on generic Linux hardware.

Should you self-host clawbox?

clawbox scores 38/100 on selfhost.directory (how scoring works) and ranks #96 of 120 in AI Assistant. It has 32 GitHub stars.

It shipped 21 stable releases in the past year. The latest, v4.2.0, came out on 5 October 2026. New versions land roughly every 5 days.

Setup is moderate: expect some configuration beyond a single command. It is Other-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is hermes-agent (89/100, 257 957 stars). It scores higher overall, so compare the two before you pick. See clawbox vs hermes-agent side by side.

Features

  • Browser Automation
  • Cloud AI providers
  • Coding agentRequires ClawBox AI Pro or Max plan
  • Desktop Environment
  • Encrypted backups
  • GitHub project backupAdded in v4.2
  • Kiosk ModeAdded in v4.2
  • Local AI inference
  • Memory ShardRequires ClawBox AI Pro or Max plan for cloud indexing; local indexing available
  • Monitor modeAdded in v4.2
  • Multi-platform Messaging
  • Multi-user supportAdded in v4.2
  • Remote Access
  • Switch model mid-chat
  • Voice input and output

clawbox README

How to install clawbox

Install clawbox with Linux

Environment variables
CLAWBOX_PORT=3005, CLAWBOX_GATEWAY_PORT=18789, CLAWBOX_TERMINAL_WS_PORT=3006, CLAWBOX_SKIP_DESKTOP_SERVICES=0

sudo bash install-x64.sh — full install sudo bash install-x64.sh --step NAME — run a single step sudo bash install-x64.sh --preflight — check the host, change nothing

Ports configurable via CLAWBOX_PORT (default 3005), CLAWBOX_GATEWAY_PORT (default 18789), CLAWBOX_TERMINAL_WS_PORT (default 3006); CLAWBOX_SKIP_DESKTOP_SERVICES=1 leaves existing VNC/websockify/browser services untouched

How to configure clawbox

  • ALLOWED_HOSTSclawbox.local,10.42.0.1,10.43.0.1,localhostTrusted hostnames

    Trusted hostnames

  • ALLOW_INSECURE_CONTROL_UItrueAllow insecure control UI origins

    Allow insecure control UI origins

  • CANONICAL_ORIGINhttp://clawbox.localDefault redirect origin

    Default redirect origin

  • CLAWBOX_AI_API_KEYEnables ClawBox AI as an automatic OpenClaw fallback model

    Enables ClawBox AI as an automatic OpenClaw fallback model

  • CLAWBOX_CONTROL_UI_ORIGINS_FILE/home/clawbox/clawbox/data/control-ui-origins.jsonExtra trusted control UI origins

    Extra trusted control UI origins

  • CLAWBOX_GATEWAY_PORT18789OpenClaw gateway port on x86_64 (default: 18789)

    OpenClaw gateway port on x86_64 (default: 18789)

  • CLAWBOX_MCP_TOKENAuto-generatedBearer token the MCP server uses on /setup-api, persisted in data/.mcp-token

    Bearer token the MCP server uses on /setup-api, persisted in data/.mcp-token

  • CLAWBOX_PORT3005Port for ClawBox UI on x86_64 (default: 3005)

    Port for ClawBox UI on x86_64 (default: 3005)

  • CLAWBOX_ROOT/home/clawbox/clawboxProject root directory

    Project root directory

  • CLAWBOX_SKIP_DESKTOP_SERVICES0Leave existing VNC, websockify and browser services untouched

    Leave existing VNC, websockify and browser services untouched

    values: 0,1

  • CLAWBOX_TERMINAL_WS_PORT3006Terminal websocket port on x86_64 (default: 3006)

    Terminal websocket port on x86_64 (default: 3006)

  • EMBED_BASE_URLhttp://127.0.0.1:8081/v1Memory-search embedder base URL

    Memory-search embedder base URL

  • EMBED_BATCH1024Embedding batch size

    Embedding batch size

  • EMBED_HF_FILEQwen3-Embedding-0.6B-Q8_0.ggufHugging Face file for embedding model

    Hugging Face file for embedding model

  • EMBED_HF_REPOQwen/Qwen3-Embedding-0.6B-GGUFHugging Face repo for embedding model

    Hugging Face repo for embedding model

  • EMBED_MODELqwen3-embedding-0.6bEmbedding model name

    Embedding model name

  • EMBED_N_GPU_LAYERS99Number of GPU layers for embedding model

    Number of GPU layers for embedding model

  • GATEWAY_PORT18789OpenClaw gateway port

    OpenClaw gateway port

  • LLAMACPP_BASE_URLhttp://127.0.0.1:8080/v1Connect the llama.cpp local provider to a running llama-server

    Connect the llama.cpp local provider to a running llama-server

  • LLAMACPP_BIN/usr/local/bin/llama-serverPath to llama-server binary

    Path to llama-server binary

  • LLAMACPP_CONTEXT_WINDOW131072llama.cpp context window size

    llama.cpp context window size

  • LLAMACPP_HF_FILEgemma-4-e2b-it-edited-q4_0.ggufHugging Face file for llama.cpp model

    Hugging Face file for llama.cpp model

  • LLAMACPP_HF_REPOgguf-org/gemma-4-e2b-it-ggufHugging Face repo for llama.cpp model

    Hugging Face repo for llama.cpp model

  • LLAMACPP_MODELgemma4-e2b-it-q4_0Bundled local model served by llama.cpp

    Bundled local model served by llama.cpp

  • NETWORK_INTERFACEwlP1p1s0WiFi interface for AP

    WiFi interface for AP

  • NOVNC_WS_PORT6080noVNC WebSocket port (remote desktop)

    noVNC WebSocket port (remote desktop)

  • OLLAMA_HOSThttp://127.0.0.1:11434Ollama server URL

    Ollama server URL

  • PORT80Web server port

    Web server port

  • SESSION_SECRETAuto-generatedSession cookie signing key, persisted in data/.session-secret

    Session cookie signing key, persisted in data/.session-secret

  • TERMINAL_WS_PORT3006Terminal WebSocket PTY port

    Terminal WebSocket PTY port

clawbox releases & version history

Even more releases RSS

FAQ

Are there any ongoing subscription fees?

ClawBox hardware is €749 one-time. The free plan includes chat, image generation and voice transcription with weekly allowances. Optional Pro (€9/mo) and Max (€49/mo) plans add Coding Agent, Memory Shard, cloud voice replies, remote access at your own clawbox.tech link and larger allowances.

Does it work with my existing smart home setup?

Yes! ClawBox integrates with Home Assistant, MQTT, REST APIs, and most popular smart home platforms. It can control lights, thermostats, security systems, and more through voice commands or automated rules.

What AI models come pre-installed?

ClawBox ships with several optimized models: Llama 3.1 8B (general conversation), CodeLlama 7B (code assistance), Hermes 3 8B (tool usage), and LLaVA 7B (vision tasks).

Can it run larger models like CodeLlama 34B?

ClawBox works best with models up to 7-8B parameters (8GB unified memory). For larger models, it uses intelligent model switching and can offload to cloud APIs when needed.

How much power does ClawBox use?

ClawBox draws just 20W under normal AI workloads — about €0.80/month in electricity (at €0.25/kWh). Even under heavy inference, it stays below 25W.

What storage does ClawBox use?

ClawBox comes with a 512GB NVMe SSD for significantly better I/O performance when loading AI models and handling database operations.

Can I run OpenClaw OS on my own Jetson?

Yes — this repository is source-available and installs on an NVIDIA Jetson Orin Nano 8GB running JetPack 6.2.

Where do I buy a ClawBox?

Only from clawbox.com. ID Robots ships to 108 countries via DHL Express. Products sold elsewhere under a similar name are not ClawBox and are not covered by warranty or support.

How do I reach my box from outside my network?

Turn on Settings → Remote Control. A provisioned box gets a persistent <boxHandle>.clawbox.tech address over a named Cloudflare tunnel. No Cloudflare account is needed. If the named credential is missing or refused, the box falls back to a quick tunnel with a random *.trycloudflare.com URL.

What is the Coding Agent?

A headless Claude Code run you hand a whole task to. It works in a git worktree copy of your project, opens a pull request with gh, and runs an automatic review pass that adversarially reviews its own diff. Runs continue in the background and you can steer them mid-flight. Requires Pro or Max plan.

Which AI model answers by default?

On ClawBox AI it is DeepSeek V4 Flash (deepseek-v4-flash) with a 1M-token context window. The Max plan uses DeepSeek 4.1. You can instead use your own Anthropic, OpenAI, Google or OpenRouter key, sign in with a subscription you already pay for, or stay entirely on the box.

Does ClawBox work without internet?

Yes, for local models. Gemma 4 E2B ships preinstalled on the box's own llama.cpp, and Local-only mode routes every request to it with cloud providers switched off. Internet is needed only for updates, messaging integrations, browser automation, and optional cloud AI providers.

What is the difference between ClawBox and OpenClaw?

OpenClaw is the open-source AI agent; ClawBox is the dedicated hardware appliance that runs it 24/7 with OpenClaw OS on top — desktop environment, setup wizard, built-in apps, backups and updates.

Who makes ClawBox?

ClawBox is made by ID Robots Ltd., a robotics and AI company based in Plovdiv, Bulgaria (EU).

More in Self-hosting

Rate this project

Click to rate — no account needed.

0 comments

Posting as guest
Loading comments…