KumoKodo runs twenty-plus products under one roof. Six months in, the bottleneck wasn't code — it was context. Decisions from product A would surface again in product C and we'd re-litigate them because the rationale was scattered across Slack threads, half-written README's, and tribal knowledge. Lessons from a 3am incident in March wouldn't carry to May. Marketing copy drifted from what the code actually did within weeks of every ship.
We tried Notion. We tried Obsidian. We tried GBrain. They all share the same shape: a passive store that waits to be queried. Useful, but only when you remember to go there. The problem with a portfolio brain isn't storing the knowledge; it's getting it back at the moment it matters — which is the moment you start writing the next file, not the moment you have time to sit down and search.
Roushi is what came out of building for that constraint. A brain that runs while you code. PreToolUse hooks inject the relevant lessons, decisions, and rules into Claude Code before each Edit/Write. Cron jobs audit the brain itself for stale, duplicate, and contradictory entries. Skills are indexed alongside rules so a single source of truth covers both what the agent should know AND when it should invoke a procedure.
The name is 老師 (rōshi — old master). The brand has a single accent: vermilion, used like a scholar's seal stamp. The metaphor is intentional. A brain isn't a database; it's the consulted wise one.
I built Roushi for myself first. KumoKodo is the dogfood; private beta is open to other portfolio operators who want the same thing. When it stabilizes, it graduates into a multi-tenant SaaS — but the operator-first design stays.
— Sam, KumoKodo