OpenClaw is a local-first personal assistant designed to do work across a computer rather than live inside one chat tab. It runs on macOS, Windows and Linux, connects to the messaging services people already use and can operate browsers, files and shell tools. The project is open source and stewarded by an independent nonprofit, which makes it unusually inspectable and adaptable.
What stands out
OpenClaw supports 29 chat channels, including WhatsApp, Telegram, Discord, Slack, Signal and iMessage. Users bring their preferred hosted, subscription, gateway or local model, then extend the assistant with skills and plugins. Persistent memory carries context between conversations, while workspace and sandbox controls help separate an agent’s access from the rest of the machine.
- Runs on a user-controlled computer with desktop apps for the major operating systems
- Works through familiar chat channels rather than a proprietary inbox
- Uses browsers, local files, command-line tools and connected services
- Supports multiple agents, scheduled work, voice and a visual canvas
- Offers skills, plugins and a public extension ecosystem
- Provides model choice instead of tying the user to one provider
Pricing
OpenClaw itself is free under the MIT licence. There is no official subscription, hosted tier or enterprise edition. Users still pay any model API, gateway or cloud-compute charges they choose to connect, and those costs can vary considerably with model choice and workload. A local model can reduce external usage fees but demands suitable hardware.
Our take
OpenClaw is compelling for technical users who want control, broad integrations and an assistant that can take action on their own machine. That power brings responsibility: installation, permissions, secrets and sandboxing need careful attention, particularly when third-party skills are involved. It is a stronger fit for people comfortable administering software than for teams seeking a fully managed service with central support and predictable per-seat billing.