Hi devs,
I keep it short.
Go runtime, Pi-ready. Typed envelopes, observable groups, shared memory, skill routing.
Docs: https://kafclaw.scalytics.io
Feedback?
best
2 Likes
I can’t understand the hype. OpenClaw is a feverish dream. Token-hungry. Immense security risk. I never would have thought that people would want this.
And by the way, not exactly hard to program either.
Anyone who should try it, definitely isolate it well and set limits. Hacks incoming…
2 Likes
It’s not a “dream”, the concept of using an orchestrator to manage isolated programs, who are able to spawn their own programs in their space, makes the use of LLMs / SLMs / own models quite comfortable. It’s the idea that counts.
Token hungry - when the user does not know what’s going on, of course. It’s the same for every tool. I use embedded vector, reduced depth, strict token awareness, local tools before LLM slop.
Re security: Honestly, as Linux came out, and till today, numerous security issues are known. Same for all software out there. The beauty of this wave of agents is open source - you find something, deliver a patch. Involve yourself, drive, and steer if you don’t like how it performs.
I said fever dream, not dream. 
Frankly, I do belive that you know what you are doing. - But my free warning is not for you. It is for people who get caught by the hype and run yolo.EXE
Ya fully agree! That’s why we do in our project a lot of skill sandboxing, hardening, external kms integration, local token encryption, security checks, and host-firewalling. And whoever installs a system like *Claw on his private or work laptop without knowing, reading docs, what shall I say 