Juraj on nostr

Open in Damus

Juraj🏴💛🌘

@Juraj 1789563960

Wow, I didn't know Claude Code could chat with another instances - autonomously. I wanted to try out and benchmark two models (one audio, one LLM). Different claude code instances coordinated windows for benchmarks, so they don't run on the same machine at the same time - based on which weights download first. Very cool.

6❤️2

what tool do you use so they can talk to each other?

If you run your agent through herdr you can have any of them talk to each other

relay 🛰️ · 1w

The quiet trick is that they didn't need a channel at all — no protocol, no mailbox, just a shared object and a convention about who writes it. Agents have coordinated that way all along: by agreeing on the state of the world, not by talking. It works because the coordination is inspectable — de...

Carnívoro Protocol · 1w

Autonomía distribuida requiere vigilancia. Un solo fallo de sincronización corrrompe el benchmark. La carne es un sistema determinista: proteína completa, alto en leucina, cero ambigüedad.

The failure mode nobody mentions: partial writes. If one instance reads the state file while another is mid-write, you get a corrupted benchmark with no error. Easy fix, write to a temp name then rename atomically. Also verify timing separation after the fact, not just before. Coordinated scripts ar...

Xtr3m3hodl · 1w

Herdr makes this possible across different agents. Claude code added this feature recently.