Chat with your local AI assistant from existing messaging apps
Send a message in a familiar messenger app, and a program running on your computer receives the request and handles the task. OpenClaw is an open-source assistant designed to keep state and memory on your hardware while allowing you to swap various AI models and tools like plugins. From a personal laptop to a shared team server, you can operate the same structure by simply changing the configuration.
The chat window is the interface
It connects to more than 20 messaging platforms, including Discord, iMessage, Slack, Teams, Telegram, and WhatsApp. You can talk to the assistant in your existing chat channels without installing a separate app, so there is no need to learn a new interface. Native apps for macOS, iOS, Android, Windows, and Linux are also provided, allowing you to use device-specific features such as voice, camera, and screen sharing. All connections are managed through a Gateway that runs locally, which serves as the central hub controlling sessions, tools, events, and channel connections.
Data sovereignty and model flexibility
State, memory, and credentials are all stored on the user's hardware. Data does not leak to cloud services, and you can freely swap model providers or agent harnesses (the framework that runs the model) with Claude, Codex, local models, and others. By default, it only performs a daily version check, and anonymous usage statistics are optional. Setting update.checkOnStart: false disables all this external communication. It is operated by the independent 501(c)(3) OpenClaw Foundation, with no paid tiers or hosting services.
Installation and initial setup
You can install it on macOS, Linux, and Windows using an installation script, which also automatically prepares the required Node.js runtime. If you manage Node.js directly, you can install it with the npm install -g openclaw@latest --allow-scripts=openclaw command. After installation, running openclaw onboard --install-daemon opens the onboarding wizard, which automatically proceeds with checking model access permissions, creating a workspace, and configuring the Gateway. Once complete, run openclaw dashboard to send a message in the Control UI and verify that it is working properly.
Security and pre-deployment checks
All received messages must be treated as untrusted input. In channels where DMs are possible, pairing with unknown senders is blocked by default and must be approved with the openclaw pairing approve <channel> <code> command. Tools run on the host system by default, so if you do not configure sandboxing, they can access the system directly. You must read the security guide, exposure runbook, and sandboxing guide before connecting other users or exposing the Gateway remotely. This repository is structured as a pnpm workspace, and npm install in the root directory is not supported.