Curator in the agent stack
Curator is the memory layer that keeps agents current about you — here's how it fits with the other four tools.
Static memory remembers what you said. RAG retrieves notes that look similar. Curator is a third thing: evolutionary memory that applies what’s changed since. Come back to a project from last year and Curator doesn’t hand the agent a pile of stale notes — it hands over the delta between the engineer who wrote it and the one you are now: the defaults you’ve shifted, the patterns you now reject, the standards you’ve raised. It’s migration history for your own judgment.
The four it works with
- TheAuditor brings facts about the code; Curator brings facts about you.
- Warden acts with both ground truth and memory already in context.
- Arbiter orchestrates the whole stack across providers.
- BenchProctor applies the same proof discipline to the SAST the stack relies on.
What Curator brings
Memory that tracks how your standards move over time — when you level up, your old projects inherit the upgrade. On returning to a codebase it surfaces the diff in your standards before the agent proposes a single edit. It’s local-first and encrypted: it runs on your machine, no cloud, no account, no telemetry. It’s MCP-native, so it works with Claude Code, Cursor, Windsurf, Codex, and Warden. Every memory lives in a named domain so contexts don’t bleed into each other — and every recall explains why it surfaced, not just that it did.
The ecosystem framing says it best: don’t let agents guess about your code, don’t let them guess about you, and let the agent act with both.
Where it stands
Curator is in early access, with the pre-alpha being built now. Join the early-access list and we’ll email when it’s ready. Background reading: the memory category nobody shipped and resume with what you know now.
Was this useful?