What Is OpenClaw (2026)? Install the Self-Hosted AI Agent
By Antigravity Skills editorial team · Last updated
Short answer: OpenClaw is an open-source AI assistant that runs on your own machine and works through messaging apps you already use, such as WhatsApp, Telegram, Slack, Discord, Teams and iMessage. Its state, memory and credentials stay on your hardware, and the AI model behind it is a swappable plugin. The code is MIT licensed. You install it yourself from the official docs; there is no hosted service or paid tier described in the repository.
Unofficial guide. Antigravity Skills is not affiliated with OpenClaw. Every fact below comes from the official GitHub repository and getting-started docs, checked on November 18, 2026. Requirements and commands change, so confirm them on the official pages before installing.
What is OpenClaw?
According to its README, OpenClaw is a personal (or shared) AI assistant you host yourself. Instead of opening a separate app, you message it in a chat app and it acts on your computer. A few pieces from the README:
- Gateway: a local service that manages sessions, tools, events and channel connections.
- Interfaces: a Control UI (dashboard), a CLI and a terminal UI.
- Channels: many messaging apps, with companion apps and device nodes for features such as voice, Canvas, camera and screen on supported platforms.
- Swappable models: providers and agent harnesses, for example Claude, Codex or local models, are plugins.
- Extensions: tools, skills and plugins, with a plugin marketplace called ClawHub.
- Safety defaults: unknown senders must be paired before they can talk to it, and tool execution can optionally be sandboxed.
The project says it is stewarded by the OpenClaw Foundation, a nonprofit.
Open source and self-installed
OpenClaw is MIT licensed. Because you run it yourself, you are responsible for the machine it runs on, the accounts you connect and the model access it uses. Install it only from the official sources named below, and read the documentation's security notes before connecting it to personal chat accounts.
How to install OpenClaw
The official getting-started page lists these options.
macOS, Linux or WSL2:
curl -fsSL https://openclaw.ai/install.sh | bashWindows PowerShell:
iwr -useb https://openclaw.ai/install.ps1 | iexTo try it without installing:
npx openclaw@latestThe docs also list Docker, Nix and npm routes on their Install page. If you already use Node.js, the README gives an npm global install; the exact flags depend on your npm version, so copy them from the README rather than from memory.
Requirements (from the docs): Node.js 24.16.0 or newer on the 24.x line, or 26.1.0 or newer. The docs recommend Node 26 and newer patch releases for security fixes, and you need AI access, either an existing Claude Code or Codex CLI login or a provider API key.
After installing
- The installer starts the onboarding wizard. Choose Quick start to reuse detected AI access, or Custom setup for the full flow.
openclaw onboard --classicopens the step-by-step wizard. - If the background Gateway service was not installed during onboarding, run
openclaw gateway installandopenclaw gateway start. - Check it with
openclaw gateway status; the docs say it should listen on port 18789. - Open the dashboard with
openclaw dashboardand send your first message in the chat.
If something fails, openclaw triage runs read-only health checks and openclaw doctor shows the findings.
What can OpenClaw do?
The README describes it as an assistant that "really does things": you message it, and it uses tools and skills on your computer through the Gateway. What it can do in practice depends on the tools, skills and plugins you enable and the model you connect. For the current list of tools and channels, use the official docs rather than a third-party summary.
OpenClaw also appears as one of the CLI-based agents in Higgsfield's help center. We cover that in Higgsfield + Claude.
FAQ
Is OpenClaw free and open source?
The repository is MIT licensed, and it states there is no paid tier or hosted service. You still pay for whatever AI model access or API keys you connect.
Where does my data live?
The README says state, memory and credentials are kept on your own hardware. Messages sent to a model provider are handled under that provider's policy.
Which models does it work with?
Model providers and agent harnesses are plugins, with Claude, Codex and local models given as examples.
Where do I get help?
The README points to docs.openclaw.ai and community.openclaw.ai.
Related on Antigravity Skills
- How to install skills in Antigravity, Claude Code and Codex
- Higgsfield + Claude: mentions OpenClaw among CLI-based agents
- gstack
- All skills