Frequently Asked Questions

OpenClaw is an open-source, self-hostable AI agent framework that lets you run a powerful autonomous assistant on your own hardware. Unlike cloud-based AI assistants, OpenClaw runs locally, giving you complete control over your data and configuration. It connects to various messaging platforms and can be extended with skills for endless possibilities.

Minimum requirements:

• 2GB RAM (4GB recommended)
• 10GB disk space
• Node.js 18+ or Docker
• Internet connection for API calls

For local LLM processing, you'll need more RAM and preferably a GPU.

The fastest way to get started is via npm:

npm install -g openclaw
openclaw init

Or with Docker:

docker run -it openclaw/openclaw

After installation, run the setup wizard to configure your first channel and API key.

Not at all! OpenClaw can be used entirely through configuration files and built-in commands. However, knowing JavaScript/TypeScript helps when creating custom skills. For most users, the documentation and skill library are sufficient.

OpenClaw supports multiple providers:

• OpenAI - GPT-4o, GPT-4 Turbo, GPT-3.5 Turbo
• Anthropic - Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 Haiku
• Google - Gemini Pro, Gemini Ultra
• Ollama - Run LLMs locally (Llama 3, Mistral, etc.)
• OpenRouter - Access 100+ models through one API
• Azure OpenAI - Enterprise deployment

OpenClaw itself is open-source and free. However, you need an LLM API key, which typically costs money based on usage. Options:

• Free tiers - Many providers offer limited free credits
• Open-source models - Use Ollama to run models locally for free
• Paid APIs - Pay only for what you use

OpenClaw supports numerous channels:

• Discord - Server bots with slash commands
• WhatsApp - QR code authentication
• Telegram - Bot API integration
• Slack - Workspace integration
• Signal - Encrypted messaging
• SMS - Via Twilio
• iMessage - Via BlueBubbles (Mac)
• Microsoft Teams - Enterprise
• Google Chat - Workspace
• Matrix - Self-hosted, E2E encrypted
• Web Chat - Custom chat widget

WhatsApp setup requires QR code authentication:

openclaw connect whatsapp

This opens a QR code in your terminal. Scan it with your WhatsApp app (Settings > Linked Devices). The session persists until you explicitly disconnect. Note: WhatsApp may limit the number of devices.

Yes! OpenClaw can listen and respond on all connected channels at once. Simply add multiple channel configurations in your gateway.yaml. Each channel maintains its own conversation context.

Your bot needs these permissions:

• Read Messages
• Send Messages
• Embed Links
• Attach Files
• Use Slash Commands

Create a bot role with these permissions and assign it to your OpenClaw bot.

Skills are modular extensions that add capabilities to your OpenClaw agent. They can:

• Add new tools (browsers, APIs, databases)
• Define custom behaviors
• Provide knowledge bases
• Automate workflows

Skills are installed via the CLI and configured in your gateway.

• ClawHub - Official marketplace with 700+ skills
• GitHub - Search for "openclaw-skill" or "openclaw-plugin"
• NPM - Search for "openclaw-skill" packages
• Community Discord - Share and discover

Most skills install with one command:

openclaw install @skill-name

Or via npm:

npm install openclaw-skill-name

Then add to your gateway config under the skills section.

Absolutely! Skills are TypeScript/JavaScript packages. Basic structure:

export default {
  name: "my-skill",
  version: "1.0.0",
  tools: [{ /* tool definitions */ }],
  init: async (claw) => { /* setup */ }
};

See the Skills Documentation for full details.

OpenClaw uses a sophisticated memory system:

• Short-term - Current conversation context
• Long-term - Vector database for semantic search
• Forgetting - Auto-delete old memories after X days

This allows your agent to remember relevant information across sessions.

• Pinecone - Managed, popular choice
• Weaviate - Open-source
• Qdrant - Open-source, Rust-based
• Chroma - Open-source, Python-first
• Milvus - Open-source, scale-out
• MongoDB Atlas - Vector search

Yes! Use the knowledge tool to ingest documents:

claw.addKnowledge("./my-docs/", {
  type: "pdf",
  recursive: true
});

Supported formats: PDF, TXT, MD, DOCX, CSV, and more.

Almost anywhere!

Cloud VPS:

• DigitalOcean - One-click app
• Linode, Vultr, Hetzner
• AWS, GCP, Azure

Serverless:

• Cloudflare Workers
• Vercel (limited)

Self-hosted:

• Home server
• Raspberry Pi (limited)
• NAS (Docker)

Options for HTTPS:

• Caddy - Auto-HTTPS, zero config
• nginx - Reverse proxy with Certbot
• Cloudflare - Proxy + free SSL
• Traefik - Docker with auto-SSL

Yes! Use a process manager:

• Systemd - Linux service
• PM2 - Node process manager
• Docker Compose - With restart policies
• Supervisor - Simple process control

Configure auto-start on boot for uninterrupted operation.

Configure voice in gateway.yaml:

voice:
  provider: "elevenlabs"
  apiKey: "${ELEVENLABS_API_KEY}"
  voiceId: "rachel"

Supported providers: ElevenLabs, OpenAI TTS, Google Cloud TTS, Azure Speech.

Talk Mode enables real-time voice conversations:

• Voice input via STT (Speech-to-Text)
• Voice output via TTS (Text-to-Speech)
• Low latency for natural conversation

Works best with ElevenLabs for natural-sounding voices.

OpenClaw gives you full control:

• Self-host - Data never leaves your server
• Local LLMs - Process everything offline
• Encryption - E2E encrypted channels (Signal, Matrix)
• API keys - Stored locally, never transmitted

Unlike cloud AI, you decide where data goes.

Best practices:

• Use environment variables for secrets
• Enable authentication on web interfaces
• Keep OpenClaw updated
• Use firewall rules
• Run in isolated network
• Regular backups of config and memory

Common fixes:

1. Check API key is valid in gateway.yaml
2. Verify Node.js version (18+)
3. Run with --verbose for debug output
4. Check port isn't already in use
5. Validate YAML syntax

Troubleshooting steps:

1. Verify channel credentials
2. Check bot has necessary permissions
3. Ensure channel is enabled in config
4. Look for errors in logs
5. Restart the gateway

Solutions:

• Upgrade your API plan
• Add delay between messages
• Use caching for repeated queries
• Switch to a provider with higher limits
• Consider local LLM for high-volume

Schedule tasks using cron expressions:

cron:
  - name: "morning-briefing"
    schedule: "0 8 * * *"
    action:
      type: "message"
      channel: "discord"
      content: "Good morning! Here's your briefing..."

Standard cron format: minute hour day month weekday

Yes! Spawn specialized subagents:

const researcher = await claw.spawn({
  agent: "researcher",
  task: "Find info about X"
});

Subagents run in parallel and can handle specialized tasks.

• Official Docs - Comprehensive guides
• Discord - Active community
• GitHub Issues - Bug reports
• Our Troubleshooting Guide - Common fixes