Repo map (L1)

Scan-first index of domains for eigen product contributors. Operators and integrators should start with Architecture overview and How memory works.

Pick a domain → read its L2 file → if it points to a conflict ID, read that row in conflicts.md.

• Ingestion — Persist thoughts: normalize, classify, embed, lexical index, Postgres + Apache AGE graph sync, activity logging. L2: ingestion.md.
• Client capture queue — Browser IndexedDB queue, serial drain, NDJSON progress UI, offline sync. L2: capture-queue.md.
• Retrieval — Hybrid search (vector + lexical + AGE graph RRF), API + MCP + QA compose. L2: retrieval.md.
• Auth and tenancy — Better Auth sessions, scoped Postgres via app.current_user_id, RLS policies. L2: auth-and-tenancy.md.
• UI surfaces — SvelteKit routes and client fetch patterns for capture, graph, chat, settings, etc. L2: ui-surfaces.md.
• Consolidation (sleep) — Nightly pg_cron → salience decay, ontology prune, graph communities, open-loop boost. L2: consolidation.md.

Supporting docs (not domain maps):

• answer-protocol.md — how the coding assistant should answer when maps and code disagree.
• maintenance.md — when to update which layer after a change.
• Embeddings DB-only boundary — vectors in Postgres only; never in MCP tools or LLM calls.