Skip to content

Ingestion Lifecycle

  1. API accepts POST /v1/ingest and records a queued job in vaultcrux.ingest_jobs.
  2. Worker claims queued jobs with FOR UPDATE SKIP LOCKED.
  3. Worker applies policy gates and PII redaction.
  4. Worker chunks content and computes checksums.
  5. Worker writes documents, chunks, and vector rows:
    • private: embeddings_768
    • commons: commons_embeddings_768
  6. Worker marks job completed or failed.

Idempotency

  • Ingest job IDs are UUID-based and job payload checksum is persisted in documents.content_checksum.
  • Re-ingest of an existing doc_id updates metadata and content checksum additively.

Copyright 2026 CueCrux