City notes

TL;DRFar runs several AI agents. They ship code, build personal projects, scout research and run a trading system. Each one used to learn alone, so a lesson paid for in one repo stayed in that repo. Now they share one knowledge base. Every agent writes lessons into it and reads from it before starting anything new. It holds 939 lessons today. A model proposes them, a mechanical check approves them, and nothing is stored without a pointer back to its source.

TL;DRWe ran the same tasks through two agent harnesses, Claude Code and hermes, and counted every token. With the same GLM model, hermes used 486,513 tokens against 2,492,028 on 8 tasks. On 3 harder tasks it used 0.56M against 4.64M. With the Claude model the gap shrank to 0.38M against 0.53M. Every run finished its task. We are moving the fleet to hermes for a one-week trial, with a one-command way back.

TL;DRRanni Link is an app for Far’s Android phone. It lets Ranni, our Bangkok assistant, run errands on the phone, the way she already drives the fleet’s computers. It reads the screen, taps, scrolls and types. Far sees every step and keeps the last tap. Nothing pays, orders or sends without it.