3 ms·

I made a secure one: https://github.com/skorokithakis/stavrobot https://github.com/skorokithakis/stavrobot Everything runs in containers (I run it on a server

by stavros 8mo ago

I made a secure one: https://github.com/skorokithakis/stavrobot https://github.com/skorokithakis/stavrobot Everything runs in containers (I run it on a server along with everything else), plugins have a permission system so eg the AI can read emails but not delete or send, etc. I really like it, I run it as my main agent and it has been extremely helpful.

croes 8mo ago
Part of the usefulness is based on the same thing that makes it so dangerous. If it can only read but not act, it’s safer but less useful.
- stavros 8mo ago
  I can't restrict OpenClaw if I don't need the extra capabilities. I can restrict this.
  - croes 8mo ago
    You restrict OpenClaw by not providing it certain credentials.
    - stavros 8mo ago
      Again, with my design you can give it fine-grained access to parts of services, which OpenClaw itself cannot do. This is just a fact.