OpenClaw — Quick reference

🦞 Open-source personal AI agent that runs on your own machine and answers you through the chat apps you already use.

MIT-licensed and self-hosted — free to run, but you bring your own LLM API key (Claude or OpenAI). Estimated ~$5–15/month in API calls for moderate personal use. First-time setup is technical (Node 24+, a host machine): expect 30–60 min, not a point-and-click tool.

Install & onboard — the exact commands

Command	What it does
`curl -fsSL https://openclaw.ai/install.sh \| bash`	All-in-one install — installs Node.js and OpenClaw in one step.
`npm i -g openclaw`	Alternative install if you already have Node.js.
`openclaw onboard`	Onboarding wizard — connects your LLM provider and your first messenger. Re-run it to add a second messenger.

After onboarding, open the messenger you connected and send your bot a message — when the reply arrives, your agent is live on your machine. Docs: docs.openclaw.ai

Connect channels — the named messengers

Choose one or more during openclaw onboard. The bridge means you never need a new app — same agent, same memory, same context across all of them.

Messenger

Discord

Messenger

Slack

Messenger

Signal

Messenger

iMessage

Choose your model

Provider	What it means
Anthropic	Cloud — Claude. Messages go to the Anthropic API; needs a valid API key.
OpenAI	Cloud — GPT. Messages go to the OpenAI API; needs a valid API key.
Local models	Agent and LLM both run on your machine — private by default, no data leaves your hardware.

Set the model during openclaw onboard; change it any time by re-running the relevant config step, then send a test message to confirm the new model is active.

Core capabilities & commands

Every capability is triggered from a plain chat message in your messenger.

🌐 Browser Control

Browse the web, fill forms, extract data from any site.
"Go to the OpenClaw GitHub page and tell me the repository name."

💻 Full System Access

Read/write files, run shell commands, execute scripts — full or sandboxed, your choice.
"Write a file to my Desktop called test-agent.txt with the text: agent test successful."

🔗 Multi-step tasks

Chain capabilities in one instruction.
"Go to [URL], find [fact], and write it to Desktop/fetched.txt."

🧠 Persistent Memory

Remembers you across sessions — builds context on your preferences and work over time.

⏰ Proactive tasks

Cron-style scheduling.
"Every Monday at 8am, summarize my open GitHub tasks and this week's calendar events."

📥 Email triage

"Summarize all unread emails from today and flag anything urgent."

Skills & integrations — ClawHub

50+ integrations available. Install community skills, or write your own. Marketplace: clawhub.ai

Named integrations

Gmail · GitHub · Google Calendar · Spotify · Obsidian · Twitter · Hue (smart home) · Browser — plus the 6 messengers above, Claude & GPT, and 50+ in total.

Try it from chat

"Check my GitHub notifications"
"Show me my last 5 unread emails"
"List my open pull requests"

Browse clawhub.ai → install a skill per its instructions (or docs.openclaw.ai) → exercise it from your messenger and confirm it returns real data from the connected service.

Tips

Privacy by design: the agent runs on your own hardware with your user account's permissions — nothing passes through a third-party cloud beyond your chosen LLM API. Pick local models to keep everything on-device.

Heads-up: voice mode is currently macOS/iOS-only (wake-word); the project is in governance transition after creator Peter Steinberger joined OpenAI — check openclaw.ai for current terms. There's no visual output by default — answers arrive as text in your messenger, so complex tables/charts need OpenClaw OS on top.