Recently browsing GitHub, I discovered that Tsinghua University, Renmin University, ModelBest, and others have jointly open-sourced a project called EdgeClaw.
It lets you deploy the ‘Little Lobster’ locally, with data never leaving your device.
The mainstream ‘Little Lobster’, OpenClaw, is deployed on your local computer, but it still needs to call cloud-based large models for inference.
Your messages, files, and tool call results are all sent to cloud APIs for processing.
Passwords and such actually pass through third-party cloud services.
EdgeClaw was created to address this very issue.
It’s based on OpenClaw but adds a set of privacy protection mechanisms and can use local models to handle tasks: data never leaves your machine, and it can run even offline.
EdgeClaw is the security-enhanced version of OpenClaw.
It has everything OpenClaw has: 20+ IM integrations, browser operations, file read/write, 50+ Skills, sub-Agent system, etc.
The core additions are just three things: