Frequently Asked Questions
Get answers to common questions about OpenClaw
Get answers to common questions about OpenClaw
This page answers common questions about OpenClaw: what it is, how to install and update, how to restart or fix “no output,” gateway token and Chrome extension errors, and how skills, memory, groups, and the Gateway work. You’ll also find system requirements, supported channels, privacy, multi-agent and multi-device setup, and where to get more help. Jump to a question below.
OpenClaw (formerly known as Clawdbot, then Moltbot) is a personal AI assistant that runs entirely on your own machine. It connects to messaging apps like WhatsApp, Telegram, Discord, Slack, Signal, and iMessage, allowing you to interact with an AI agent through the apps you already use. Your data stays private because everything runs locally on your device.
OpenClaw is an open-source, self-hosted AI assistant framework that gives you complete control over your data and AI interactions.
Yes. One OpenClaw Gateway can run many sessions—each session is effectively one “agent” with its own identity, memory, and role. You give each agent a distinct session key (e.g. agent:researcher:main, agent:writer:main), optionally a SOUL or role file in the workspace, and use cron jobs to wake them on a schedule (heartbeats). A shared task list (Notion, Convex, spreadsheet, or files) lets them coordinate. For a full walkthrough, see Multi-Agent Team use case.
To restart OpenClaw Gateway, use one of these commands:
openclaw restart – quick restartopenclaw gateway restart – explicit gateway restartIf OpenClaw is running as a system service:
systemctl --user restart openclaw-gatewaylaunchctl unload ~/Library/LaunchAgents/ai.openclaw.gateway.plist then start againFor more, see the Troubleshooting guide and CLI Reference.
If OpenClaw shows no output or stops responding:
openclaw gateway statusopenclaw gateway restart or openclaw restartopenclaw doctor --fix to fix common config issuesopenclaw channels listAgent replies in chat but never reads files, runs commands, or uses the browser? That usually means tools are disabled. Set "tools": { "profile": "full" } in your config. See Agent replies but doesn't run tools and Troubleshooting – No output.
Full guide: Troubleshooting.
This means the Control UI or client isn’t authenticated. Fix it by:
To get a token: run openclaw doctor --generate-gateway-token, then open the tokenized URL with openclaw dashboard --no-open. Paste that URL (or the token) into the Control UI connection settings.
Full steps: Troubleshooting – Gateway token missing (1008).
OpenClaw can run with tools.profile: "messaging", which only allows the agent to send messages—no file access, exec, or browser. If your agent chats but never runs tools, add "tools": { "profile": "full" } to your config and restart the Gateway. For approval-free commands (e.g. on Telegram), add "exec": { "security": "full", "ask": "off" } under tools. Troubleshooting – No output, Configuration.
Workspace files have a per-file character limit; if exceeded, the middle is dropped with no warning. Put important rules at the top. In chat, type /context list—if you see TRUNCATED, the file was cut. Increase bootstrapMaxChars and bootstrapTotalMaxChars in config if needed. Context guide.
Run openclaw backup create (optionally --only-config or --no-include-workspace). Verify with openclaw backup verify <archive>. See Workspace Backup & Restore for manual backup and restore.
Reinstall OpenClaw so the bundled Chrome extension is restored:
npm install -g openclaw@latest or pnpm add -g openclaw@latestopenclaw gateway restartopenclaw onboard --install-daemon then restartSee Troubleshooting for “Bundled Chrome extension is missing”.
openclaw doctor runs health checks on your OpenClaw setup (config, gateway, channels, workspace, permissions). With --fix, it tries to auto-fix common issues: tighten permissions on ~/.openclaw, create missing directories, fix gateway service path, set gateway.mode if unset, and similar. Run it when something breaks or after an upgrade. It does not change your API keys or channel credentials. Full list of checks: Troubleshooting – Health Checks.
To see all installed skills, use:
openclaw skills listThis will show all skills in your workspace (~/clawd/skills/). You can also browse skills directly in the file system or use ClawHub to search and install new skills.
Learn more in our Skills Guide.
The project has evolved through several names: it started as Clawdbot, then was rebranded to Moltbot (combining "CLAW" and "TARDIS" - because every space lobster needs a time-and-space machine), and is now known as OpenClaw. Despite the name changes, the core mission of providing a privacy-first, self-hosted AI assistant has remained constant. The CLI command and some configuration files may still reference older names for backward compatibility, but the project is now officially called OpenClaw.
OpenClaw requires:
The one-liner installer will automatically install Node.js if you don't have it.
Yes! OpenClaw runs entirely on your machine. Your conversations, memories, and configuration are stored locally as folders and Markdown files. The only external connection is to your chosen LLM provider (like Anthropic or OpenAI) for processing. Your data never leaves your control unless you explicitly configure it to.
For additional security, OpenClaw supports sandboxing for group chats and untrusted DMs, and you can configure strict DM policies requiring pairing before allowing access. Security guide.
Yes, with proper configuration. OpenClaw has security controls enabled by default: DM policy defaults to pairing (unknown senders need an expiring code); exec security defaults to deny with approval for dangerous commands; allowFrom defaults to self-only when not set; gateway auth is required. Run openclaw security audit --deep to verify your setup. The project runs a formal Trust & Security Program (threat model, roadmap, code review, vulnerability triage). See Security guide and trust.openclaw.ai.
Report privately—do not open a public issue. Report to the repo where the issue lives: Core CLI, Gateway, macOS/iOS/Android apps → openclaw/openclaw Security; ClawHub → openclaw/clawhub; Trust/threat model → openclaw/trust. If unsure, email security@openclaw.ai. Include severity, impact, reproduction steps, and remediation advice. The Trust program defines response SLAs (e.g. critical: 24h first response, 7-day fix target). Good-faith researchers are not pursued legally. Full requirements and out-of-scope: SECURITY.md. Security guide.
OpenClaw supports:
And many more via plugins and extensions. Channels.
Yes, you'll need access to an LLM provider. OpenClaw supports:
You can use OAuth for subscription-based access or API keys. OpenClaw itself is completely free and open source.
Use tiered model routing: reserve your best model (e.g. Claude Opus) for important conversations and decisions; use cheaper, faster models for heartbeats, cron jobs, and background work. Configure failover in Configuration and see Model Providers and Example Setups & Model Routing for patterns. Some users also use third-party plugins that route each request automatically to the cheapest model that can handle it—for example ClawRouter (smart router, 41+ models, pay-per-request with USDC). We don’t endorse or guarantee third-party tools; see Skills for discovery.
Often it’s because one premium model is used for everything—main chat, heartbeats, subagents, and simple tasks—so you pay top-tier pricing for work that a cheaper model could handle. Use tiered routing: reserve the expensive model for important conversations and decisions; use a cheaper, fast model for scheduled checks and background work. See Model Providers and Example Setups & Model Routing for patterns, and How can I reduce my API or model costs? for concrete steps.
The easiest way is using the one-liner installer:
curl -fsSL https://openclaw.ai/install.sh | bashThen run the onboarding wizard:
openclaw onboard --install-daemonFor detailed instructions, see our complete installation guide. You can also use the Setup Wizard in your browser to get a ready-to-run install script for your OS, then configure and add channels or skills from the same flow.
The Gateway is OpenClaw's control plane - a single long-running process that manages all channel connections and the WebSocket control interface. It runs on port 18789 by default (loopback-only for security). The Gateway handles:
You typically run one Gateway per host. It can be installed as a system service so it runs automatically. Architecture.
Yes! OpenClaw supports group chats with several safety features:
You can configure group policies per channel to control access and behavior. Channels, Configuration.
Skills are extensions that add functionality to OpenClaw. They can:
Skills are stored as Markdown files in your workspace (~/clawd/skills/). You can:
npx clawdhub@latest install <skill-name>Learn more in our Skills guide.
OpenClaw stores memories as Markdown files in your workspace. It automatically creates daily notes and remembers context across conversations. Memories are:
The agent uses these memories to maintain context and remember your preferences, making it uniquely yours over time.
Yes! You can:
Each Gateway instance manages its own channels and sessions. You can route different channels to different agents or share sessions across devices.
Yes! OpenClaw includes a browser-based Control UI (Dashboard) that provides:
Access it at http://127.0.0.1:18789/ when the Gateway is running locally. For remote access, see our remote access guide.
To update OpenClaw:
openclaw updateAfter updating, run openclaw doctor to check for issues and verify your configuration. You can switch between release channels (stable, beta, dev) using:
openclaw update --channel stable|beta|devIf you’re having problems:
openclaw doctor to diagnose issuesMost issues can be resolved by checking the configuration or reviewing the logs.
Yes! OpenClaw is completely open source under the MIT License. You can:
OpenClaw is becoming an independent foundation and remains open source; the project is maintained by the community, with ongoing support for the open-source project.