Hey HN, I built this because I liked the idea of agents controlling a computer (like Clawdbot), but I was uncomfortable giving them unrestricted shell access.
So I build Chord, it uses the same underlying agent framework as Clawdbot, so it can do most of the same jobs. The key difference is that I added a security layer. Commands are analyzed by an AI before execution. This does use extra tokens, but you can run the analysis with cheap models (e.g. Gemini Flash) to keep costs low.
The security layer can block risky commands and provides a security audit log of everything the agent tries to do.
Right now, Chord only supports Telegram as the control interface. I’ve built all the core parts, and I’ll add smaller features next.
So I build Chord, it uses the same underlying agent framework as Clawdbot, so it can do most of the same jobs. The key difference is that I added a security layer. Commands are analyzed by an AI before execution. This does use extra tokens, but you can run the analysis with cheap models (e.g. Gemini Flash) to keep costs low.
The security layer can block risky commands and provides a security audit log of everything the agent tries to do.
Right now, Chord only supports Telegram as the control interface. I’ve built all the core parts, and I’ll add smaller features next.
I’d really appreciate any feedback.