OpenClaw

OpenClaw is an MIT-licensed personal AI assistant that users run on their own computer or server. A central Gateway connects models, tools, sessions, events, messaging channels, and optional companion apps. The project is designed primarily for a single operator who wants an assistant available through chat services while retaining control of its workspace and configuration. It is distributed as the openclaw npm package, with guided installers for macOS, Linux, WSL2, and Windows.

The software itself is free and open source, but operation normally requires access to a supported model provider through an API key or supported subscription login. Messaging services, speech providers, remote hosting, and other integrations can introduce their own costs. OpenClaw is not automatically isolated: its README says main-session tools run on the host unless sandboxing is configured, so channel allowlists, sender pairing, credential handling, and remote-exposure guidance matter in real deployments.

Top features

• Local-first Gateway: one control plane manages sessions, tools, events, channel connections, and the web Control UI.
• Messaging reach: documented plugins connect services including WhatsApp, Telegram, Slack, Discord, Signal, Google Chat, Matrix, Teams, and others.
• Multi-agent routing: accounts, channels, and peers can route to agents with separate workspaces, state directories, and histories.
• Tools and extensions: skills, plugins, webhooks, scheduled work, browser tooling, and model-provider integrations expand the assistant.
• Companion apps and nodes: platform nodes can expose device-local capabilities such as voice, Canvas, camera, screen, and supported actions.
• Operational controls: onboarding, doctor commands, readiness checks, pairing, allowlists, sandbox documentation, and security runbooks support administration.
• Cross-platform setup: official onboarding covers macOS, Linux, WSL2, and Windows, with Docker and other deployment paths documented.

Use cases

• Run a self-hosted assistant on a workstation or server and reach it from multiple chat apps.
• Separate work, personal, and experimental conversations into routed agents with independent state.
• Trigger scheduled summaries, reminders, repository checks, or personal automations from chat.
• Connect an assistant to a browser or device node while keeping the Gateway under the operator’s control.
• Build a plugin-backed assistant for a team prototype, with explicit sender and tool restrictions.