
Review · AI agents
OpenClaw review: pricing, strengths and weaknesses
Technical users hosting a personal agent and managing its permissions, models and security.
Open source, freeAI agents
The score in detail
Our scoring gridPriceWeight 25 %
4.0Free plan and trialWeight 15 %
5.0Ease of useWeight 20 %
2.0FeaturesWeight 25 %
5.0Data and GDPRWeight 15 %
4.0Plans and prices
Open source (MIT)$0free software; hosting and AI model calls billed by third parties
MIT license. An assistant that runs on your computer or server, connected to 20+ messaging apps (WhatsApp, Telegram, Slack, Discord, iMessage, Signal), apps for macOS, iOS, Android, Windows and Linux, hosted or local models. Per the publisher: no paid tier and no hosted service.
Free plan: Fully free software under the MIT license, with no paid edition. Calls to the AI model you pick (your own API key) and any server you rent are paid to third parties.
Strengths
- Free under the MIT license, with no paid tier
- Controlled from 20+ messaging apps, including WhatsApp and Telegram
- Memory and credentials stored on your own machine
Weaknesses
- By default, tools run directly on the host machine, with no sandbox
- Incoming messages can carry prompt injection: security is yours to configure
- AI model costs are paid separately, with your own API keys
Data and GDPR
- Data hosting
- On your own hardware: state, memory and credentials stay on the machine running OpenClaw (the ~/.openclaw folder). Messages processed by the AI go to the model provider you pick, unless you use a local model.
- GDPR
- No GDPR policy or DPA published: the publisher runs no hosted service. By default only a daily version check is sent to the foundation (version, OS, Node.js, architecture), and it can be turned off; anonymous statistics are opt-in. Security audit carried out by Trail of Bits.
- Self-hosting
- Yes
