OpenClaw review

OpenClaw is open-source personal-agent software that you install and run yourself, rather than a hosted API or subscription SaaS. It sits in your application stack as a self-hosted agent: the desktop apps bundle the gateway, chat, setup and node features, and the homepage also offers command-line and source-code installation. OpenClaw was called Clawdbot until January 2026 and briefly Moltbot before taking the OpenClaw name.

The project targets people who want a personal agent running on their own macOS, Windows or Linux machine. Its desktop apps install Node.js and the other components for you; command-line users can install with npm, pnpm or the one-line installer. Developers can also clone the source repository and install its pnpm workspace. That makes OpenClaw a software layer to deploy and maintain, not a managed service with a vendor operating the agent for you.

The homepage's documented developer surface is installation and operation through the CLI, plus a source checkout. It gives commands for the one-line installer, npm i -g openclaw, openclaw onboard, and a pnpm workspace setup from the Git repository. To run from source, it instructs users to enable Corepack, install dependencies with pnpm and run pnpm openclaw onboard.

The homepage does not establish an API contract or name SDKs or supported programming languages. Treat OpenClaw as an installable agent rather than assuming it offers a hosted REST API, webhooks or language SDKs. Confirm API, SDK and integration support in the project's documentation before designing around one. The homepage also lists openclaw update --channel dev and openclaw update --channel stable to switch update channels. Desktop apps are offered for macOS 15+, Windows 10/11 and Linux, with architecture and package options listed for each.

The listed plan is OpenClaw (Open Source / Self-Hosted): $0 per month under the MIT License. There is no hosted tier, paid subscription or annual billing. Seat and usage limits are not published, and the pricing information lists no add-ons. That makes the OpenClaw software free, but it does not put a ceiling on the total cost of running an agent.

You supply and operate the host yourself. If your setup uses paid model or API services, those providers may add usage charges; Reddit users report that API use can become expensive, but there is no OpenClaw-specific rate or usage price to calculate from the published plan. The vendor does not publish request limits, overage rates or a production usage schedule. Before deploying, estimate infrastructure and any external model costs separately, and verify their billing and limits with those providers. OpenClaw's $0 price is most useful when you can handle self-hosting and keep the rest of the stack within budget.

OpenClaw is available under the MIT License, and the homepage links to the source repository and gives instructions to clone it and install the pnpm workspace. You can run it yourself on macOS, Windows or Linux; there is no hosted OpenClaw tier to depend on. This offers control over where the software runs, but also makes deployment, updates and security your responsibility.

The homepage does not describe a formal data-export process or document migration steps. Source access gives developers the option to inspect or fork the software, but it does not by itself guarantee that agent state, configuration or external service connections move cleanly to another tool. Check how your data and configuration are stored before relying on them, and keep copies you can restore or migrate.

The vendor does not publish seat or usage limits, request-rate limits, service-level commitments or production reliability figures. Since OpenClaw is self-hosted, you need to plan for your own uptime, upgrades, access controls, backups and incident response. The homepage lists desktop operating systems and update channels, but those facts are not a substitute for a security or compliance review. Evaluate permissions and exposure before connecting an agent to sensitive accounts or systems.

Reddit accounts are mixed. Some users describe building useful workflows, while others say their agent behaved like a chatbot, broke configuration or took time and money to make useful. One commenter describes restricting an agent's access and using extra checks for sensitive actions; that is an individual setup, not a built-in security guarantee. Treat these reports as reasons to test your own workflow, limit permissions and keep consequential actions reviewable before depending on it.