Brain · Memory moat
Memory moat
What this org has learned, and what it cost to learn it. Each finished mission (a graph run your agents already produced) folds into one durable lesson that carries its cost, so the brain reads as a priced asset that grows per run. No new capture.
Where to find it
- Localhost:
/memory-moat.html?repo=<id> - API:
GET /api/memory-moat?repo=<id> - Keyboard: ⌘ K then
memory moat - Sidebar: Team brain → Memory moat
What it does for you
Four numbers at the top.Lessons learned (and how many came from a mission), cost to learn it, recurrences prevented, recurrences observed. The cost is the priced-lesson total; a mission with no priced node reads n/a, never $0.
One row per finished mission.Each row shows what the org learned, the mission outcome (completed or failed), the cost that lesson carries, and how many times it has since caught a repeat. Cost is the mission's own rollup, honest-null when nothing in the run was priced.
Themes across your brain.The nightly dream pass groups your local brain by record type (errors and incidents, decisions, sessions, features) and writes a short theme per region, each citing the records it summarized. A theme reads "model" when your Anthropic key wrote it, "local" when the no-LLM distiller did at no cost. Regions with too little data are left out, not guessed.
How the fold works
A lesson folds from a mission via lib/mission-lessons.mjs, dedup-guarded on the mission id so a re-observed run adds no duplicate. The cost each lesson carries is the mission rollup (rollupMissionCost). It is a read over the same lessons the Lessons tab shows and the same spend Telemetry already captured. RepoOps runs no graph; it remembers and prices the runs your agents produce.