โ FAQ
Common questions about OpenClaw, answered.
๐ Getting Started
What is OpenClaw?
OpenClaw is an open-source AI agent platform that runs locally on your machine or server. It connects to messaging apps (Telegram, WhatsApp, Discord, Slack, Matrix, iMessage, Signal, QQ Bot, and more) and lets you interact with AI agents from your pocket. Think of it as a bridge between your favorite chat apps and AI assistants.
What do I need to run OpenClaw?
You need:
- Node 24 recommended or Node 22.19+ minimum (the installer handles this automatically)
- Model access via API key, OAuth/Codex, Claude CLI, OpenRouter, Ollama/LM Studio, or another supported provider
- A local computer, mini PC, VPS, or server. Use strict tool policy and allowlists if it can access private data.
Is OpenClaw free?
OpenClaw itself is 100% free and open source (MIT license). However, you'll pay for the AI API usage (e.g., Anthropic charges per token for Claude). Running costs depend on how much you use it โ typically $5-50/month for personal use.
Does my data leave my computer?
Your data is sent to your chosen AI provider (Anthropic, OpenAI, etc.) for processing. But OpenClaw runs locally โ there's no OpenClaw cloud server. Your files, memory, and configuration stay on your machine.
๐ Is it safe to run OpenClaw?
OpenClaw is powerful โ it has access to your files, shell, and network. That power requires responsibility:
- Prefer a dedicated machine or profile for high-risk automation; if you run locally, keep browser/tool access scoped
- Don't store sensitive credentials on the OpenClaw machine
- Use scoped API keys with minimal permissions
- Configure channel allowlists to restrict who can interact
โ๏ธ Configuration
Where is the config file?
The main config lives at ~/.openclaw/openclaw.json. You can edit it directly or use the configure command:
openclaw configureHow do I switch AI models?
Run the configure command and select a new model:
openclaw configure --section modelOr switch live in chat with /model. Config uses agents.defaults.model.primary and model allowlists/fallbacks.
Can I use multiple AI providers?
Yes! You can configure multiple providers and switch between them. You can also route different channels or users to different models using multi-agent routing.
What are workspace files (SOUL.md, USER.md)?
These are markdown files in ~/.openclaw/workspace/ that define your agent's personality and context:
- SOUL.md โ Core values and how the agent thinks
- IDENTITY.md โ Name, role, and capabilities
- USER.md โ Info about you (timezone, preferences)
- MEMORY.md โ Long-term memory and context
- AGENTS.md โ Guidelines for agent behavior
๐ฑ Channels & Messaging
Which messaging apps are supported?
Supported channels include Telegram, WhatsApp, Discord, Slack, Matrix, Mattermost, Microsoft Teams, Google Chat, iMessage, Signal, QQ Bot, Feishu/Lark, LINE, IRC, Twitch, Zalo, Synology Chat, WebChat, and more. See the Channels page for setup guides.
Can I use multiple channels at once?
Yes! OpenClaw can run all channels simultaneously. You can message from Telegram and get responses on Discord โ your agent handles routing.
How do group chats work?
By default, the agent only responds when mentioned (@yourbot). You can change this to always respond or configure custom mention patterns.
๐ง Troubleshooting
The gateway won't start
Common fixes:
- Check if another instance is running:
openclaw gateway status - Check for port conflicts (default: 18789)
- Run
openclaw doctorto diagnose issues - Check logs:
~/.openclaw/logs/
WhatsApp keeps disconnecting
WhatsApp Web sessions can expire. Try:
- Re-pair with
openclaw channels login - Keep your phone connected to the internet
- Don't use WhatsApp Web in your browser simultaneously
Agent responses are slow
Response time depends on your AI provider. Try:
- Use a faster/current model or lower thinking level for routine turns
- Check your internet connection
- Reduce context size by clearing old memory
How do I update OpenClaw?
openclaw updatePackage installs use the safe staged updater and restart verification. Use openclaw update status to check your channel.
๐ก Best Practices
How do I make my agent smarter?
The key is good prompting in your workspace files:
- Write a detailed SOUL.md with clear values
- Give context in USER.md about your work and preferences
- Use MEMORY.md to store important ongoing context
- Define clear anti-patterns (what NOT to do)
Should I run OpenClaw on a server?
You can, but most people run it on a Mac mini, always-on laptop, or home server. For cloud hosting, any VPS that can run the current Node line works. Tailscale makes remote access easy.
How do I backup my agent?
Backup the entire ~/.openclaw/ directory. This includes your config, workspace files, memory, and sessions. Git-track your workspace for version control.
Still have questions?
Join the community for help and discussions.