Frequently AskedQuestions

OpenClaw is an open-source AI assistant framework that runs on your own hardware. Unlike ChatGPT, it can browse the web, execute terminal commands, edit files, schedule automations, and integrate with tools like Telegram — all while keeping your data private. Think of it as a personal AI that actually lives on your machine and can do things, not just chat.

OpenClaw itself is completely free and open-source (MIT license). You only pay for the AI models you use — typically $10–100/month depending on usage. You can also run local models via Ollama completely free. See our model comparison for cost breakdowns.

Minimum: any modern computer, Raspberry Pi 4+, or a VPS with 2GB RAM and 20GB storage. Recommended: a Mac Mini M4 ($599 one-time) or a Hetzner VPS (€4.15/month) for 24/7 operation. For running local AI models, you'll want 8GB+ RAM. See the VPS hosting guide for server options.

Yes — a VPS is actually the best setup for most people. It runs 24/7 without you leaving your laptop on, costs $6–$24/month, and you get a clean Linux environment. DigitalOcean is our top pick — best docs, easiest setup, starts at $6/mo. Full details at our VPS guide.

OpenClaw supports all major models: Claude (Sonnet, Opus, Haiku), GPT-5 and GPT-5 Mini, Google Gemini 3.1 Pro, Grok 4, Kimi K2.5, and local models via Ollama (Llama 3.2, Mistral, DeepSeek, etc.). You can route different tasks to different models — cheap model for drafts, powerful model for complex work. See the full model comparison.

macOS: brew install openclaw

Linux/VPS: curl -fsSL https://get.openclaw.ai | bash

After installation, run openclaw wizard for guided configuration. The full 20-phase walkthrough is in the Setup Guide.

1. Message @BotFather on Telegram and run /newbot to create your bot.
2. Copy the bot token (looks like 123456:ABCdef...).
3. Get your user ID from @userinfobot.
4. Add both to your OpenClaw config.
5. Critical: run /setprivacy via BotFather and set it to Disabled.

More details in Phase 6 of the Setup Guide.

Add keys as environment variables in your shell profile:

export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
Then run source ~/.zshrc to reload. Never put API keys directly in config files that might be committed to git. Set spending limits at your provider dashboards.

Yes. OpenClaw supports routing different tasks to different models. You can set a default model (e.g., Claude Sonnet for most tasks), a fast/cheap model (GPT-5 Mini for quick queries), and local models as fallbacks. Configure model routing in your openclaw.json config. See model comparison for which model fits which task.

macOS (Homebrew): brew upgrade openclaw

npm (Linux/VPS): npm update -g openclaw

After upgrading, restart the gateway: openclaw gateway restart. Skills update separately via openclaw skill update --all.

Key differences: OpenClaw can run terminal commands on your machine, read/write files on your system, schedule cron jobs that fire automatically, send Telegram messages proactively, use multiple AI models interchangeably, work offline with local models, and operate 24/7 without you being present. ChatGPT can only chat in a browser tab — OpenClaw is a real autonomous agent.

Skills are plugins that give your AI new capabilities. Each skill is a folder with a SKILL.md file that tells the AI how to use it, plus any scripts it needs. Install skills via ClawHub: clawhub install weather. Over 5,400 skills are available — from GitHub integration to image generation to database queries. You can also write custom skills.

Absolutely. This is one of OpenClaw's biggest strengths. You can build: daily morning briefings, price monitors, email summarizers, GitHub PR watchers, crypto alerts, web scrapers, and much more. See 50+ real automation examples. No coding required for most automations — just describe what you want.

Cron jobs let you schedule tasks that run automatically. Add one with: openclaw cron add --schedule "0 8 * * *" --command "Send morning briefing". The AI will execute your instruction at 8 AM daily. Use standard cron syntax for timing. View all scheduled tasks with openclaw cron list.

Yes. OpenClaw runs entirely on your hardware. Your conversations, files, and automations never leave your machine unless you explicitly use a cloud AI model (like Anthropic or OpenAI), in which case only the necessary message context is sent. Use local models via Ollama for complete privacy — nothing leaves your network at all.

OpenClaw itself: $0.
AI models: $10–100/month for typical use.
VPS (optional): €4–$24/month.

Light user (few tasks/day): ~$10–20/month total. Regular user: ~$50–80/month. Power user with lots of automations: $150–300/month. Use GPT-5 Mini or local models to cut costs significantly. See model pricing.

Cheapest cloud model: GPT-5 Mini at ~$0.15/M input tokens — great for simple tasks and quick Q&A.

Best value: Kimi K2.5 at ~$0.50–1/M tokens with a 256K context window.

Free: Local models via Ollama (Llama 3.2, Mistral) — zero API cost, runs on your hardware. Check the full comparison table.

Yes. Install Ollama and pull any compatible model. Llama 3.2 (3B) is great for most tasks and runs on 4GB+ RAM. Llama 3.1 (8B) is more capable and needs 8GB+ RAM. Setup: curl -fsSL https://ollama.ai/install.sh | sh && ollama pull llama3.2. Then configure it as your default model in OpenClaw. Zero ongoing costs.

Set hard limits at your AI provider dashboards:
• Anthropic: console.anthropic.com → Billing → Usage limits
• OpenAI: platform.openai.com → Billing → Usage limits

We recommend starting with a $50/month cap and adjusting based on actual usage. You can also configure OpenClaw to prefer cheaper models by default, using expensive models only for complex tasks.

Vibe coding means building software by describing what you want in plain language, and letting AI write the code. With OpenClaw, you can say "build me a landing page for my product" and get a working HTML file. It's not magic — it works best with clear descriptions and iteration. See the full vibe coding guide.

Yes. OpenRouter is a meta-provider that gives you access to 200+ models through a single API key. This means you can use Claude, GPT-5, Gemini, Mistral, and many more without multiple API keys. Add your OpenRouter key to OpenClaw config and set the model to openrouter/auto for automatic model selection. See model comparison for more.

Your entire setup lives in ~/.openclaw/. Back this up with: tar -czf openclaw-backup.tar.gz ~/.openclaw/. Exclude API keys from backups or encrypt the archive. You can also use git for the workspace directory, just make sure .env and credential files are in your .gitignore.

Yes, with caveats. A Raspberry Pi 5 (4GB RAM) can run OpenClaw with cloud models (Claude, GPT-5) without issues — the Pi just sends API requests. Running local AI models locally is challenging: small models (1B–3B) can run but slowly. For serious use, a Pi is best paired with cloud models or used as a lightweight scheduler. A €4/month VPS gives much better performance.