Seeded from your git history
A fresh memory knows nothing about a project that already has years of history. Curator can backfill from your git log and docs overnight, so it starts useful instead of empty.
The awkward thing about giving an AI agent memory is the first day, when it has none. Your project already has years of decisions behind it, and a blank memory means re-explaining all of them before the agent is any use. Curator closes that gap by starting from what you already wrote down.
It reads the record you already have
Point Curator at a repository and it can backfill from your git history and your project documents, overnight, on its own. Your commits are a running log of what changed and, often, why. Your handoff notes and design docs are the reasoning behind the decisions. Curator turns that existing record into memory, so on day one the agent can already reference the rationale behind a decision from months ago without you narrating it again.
Safe to run, and safe to re-run
The backfill is built to be dull in the best way. It is idempotent, so running it twice does not create duplicates, and it is resumable, so if it is interrupted partway you just run it again and it picks up where it left off. It does not overwrite anything or invent history; it reads what is there. You can point it at a large project and walk away.
What you get the next morning
Once the history is in, the agent stops starting from zero. It can tell you that the pattern you are about to reach for is the one you decided against in the spring, and why. It inherits the standards you set earlier instead of relearning them. The context that used to live only in your head, or scattered across a hundred commit messages, is now something it can actually use.
Why it matters
Memory that starts empty is a memory you have to feed for weeks before it earns its keep. Memory seeded from your own history earns it on the first prompt. The work is already written down. Curator just makes the agent able to remember it.
Curator is v0.3.0 and a working pre-release product. It runs locally, is reached over MCP, and pairs with TheAuditor and the rest of Code Reality Labs. Follow release preparation.
Was this useful?