OpenClaw: the open-source personal assistant that runs on your devices

The problem with hosted AI assistants is not capability, it is tenancy: conversations, memory and credentials live at the vendor, and the model can be swapped out from under you. OpenClaw is the most-starred open-source personal assistant on GitHub (TypeScript, MIT, about 392k stars as of 2026-10-15): a Gateway running on your own machine that brings the assistant into the channels you already use — Discord, iMessage, Slack, Teams, Telegram, WhatsApp and 20+ more, plus native apps for macOS, iOS, Android, Windows and Linux.

Core features

• One Gateway, every channel: the Gateway is the local control plane for sessions, tools, events and channel connections; the Control UI, CLI and TUI all connect to it. The same config is a personal assistant on a laptop; a shared team deployment differs only in configuration.
• Models and harnesses are plugins: Claude, Codex and local models all plug in, and switching models changes nothing else. Tools, skills and plugins extend the assistant, with new capabilities distributed through ClawHub.
• Data stays on your hardware: the README states plainly that state, memory and credentials live on your machine. By default only a daily version check goes out, anonymous statistics are opt-in, and update.checkOnStart: false disables both. A dedicated docs page lists exactly what OpenClaw sends.
• Foundation-stewarded, no paid tier: developed by the independent 501(c)(3) OpenClaw Foundation, which employs the core team and signs releases. OpenAI is a donor, not an owner; there is no paid tier, hosted service or token.
• Channel security is conservative by default: the README says to treat inbound messages as untrusted input — DM-capable channels pair unknown senders by default, approved with openclaw pairing approve <channel> <code>.
• Companion apps add device capabilities: voice, Canvas, camera, screen and device-local actions on supported platforms.

Typical use cases

• Anyone who wants an assistant inside WhatsApp or iMessage without handing the transcript to a provider: run the Gateway yourself and use the channels you already have.
• Teams tired of being held hostage by a product’s model choice: models and harnesses are plugins, so Claude today and a local model tomorrow is a config change.
• Developers building a shared team assistant: one Gateway plus a team configuration, no waiting for an official enterprise edition.

Quick start

# macOS / Linux / WSL2
curl -fsSL https://openclaw.ai/install.sh | bash

Already managing Node.js? Install the package directly (Node 24.16+ or 26.1+, with 26 recommended):

npm install -g openclaw@latest --allow-scripts=openclaw

Then run openclaw onboard --install-daemon to finish onboarding (verify models, create the workspace, configure the Gateway), and openclaw dashboard opens the Control UI — send a message there to confirm the assistant works.

Summary

For people who treat “data on my own machine” as a hard requirement and still want an assistant across every channel; not for anyone who wants zero operations — the Gateway runs, and updates, on you. TypeScript, MIT, 392,069 stars as of 2026-10-15 with pushes the same day — development is very active.

Two things to know first: tools in the main session run on the host by default, and the README requires reading the security and sandboxing guides before connecting other users or exposing the Gateway remotely; and the project is funded by the foundation and its donors (Amazon, Red Hat, NVIDIA, OpenAI and others, all listed in the README). The enterprise-side control plane is the sister project OpenClaw Enterprise — a different positioning we covered at its announcement.