Knowledge Layer
Take the shortest knowledge exit first. The contract section below only matters when the map or API route still is not enough.
- Retrieval surface: Agents still reach the knowledge layer through memory_search and memory_get while Atlas and API Reference stay human-facing explanation surfaces.
- Freshness path: @lambda-curry/currykit-rag-sync still feeds Recall/QMD on a 15-minute cadence, and Almanac still handles documentation drift as a separate accuracy loop.
- Boundary rule: This landing explains the current model first, then hands off detailed topology to Network and endpoint specifics to API Reference.
Choose the shortest knowledge question first
Keep retrieval, ingestion, stewardship, and helper surfaces separate
This is the shared retrieval service. It owns indexed search, semantic recall, and the passages agents read back through memory tools.
Stay on this district when you need the shortest boundary between the retrieval surface and the pages that explain it.
The ingest pipeline still pulls published source material from Notion, Jira, and Linear into the shared Recall/QMD store on a scheduled cadence.
Atlas keeps the cadence fact here so readers can separate retrieval from the upstream feed that keeps it current.
Almanac reviews published docs for drift so the knowledge layer stays trustworthy, not merely recent.
Use this boundary when a reader starts to collapse shared knowledge stewardship into retrieval or API behavior.
The scoped helper surface on port 18790 belongs to the infrastructure story, not the default reader start path.
Leave for Infrastructure once the real question is hosting, port mapping, or the helper-surface placement around shared knowledge.