RAG Staleness Check - Read-only health checks for vector indexes
Show HN: Find stale, orphaned, deleted-but-retrievable RAG vectors
RAG Staleness Check is an open-source tool that audits your pgvector, Qdrant, or Chroma vector database for stale, orphaned, duplicate, and deleted-but-retrievable vectors. It runs read-only against your existing index, using a manifest of current documents to identify chunks whose source has changed or disappeared. It also detects near-duplicate embeddings and probes whether supposedly deleted vectors are still physically retrievable or surface in search results. This helps maintain RAG accuracy and compliance with data erasure regulations. The tool is safe to run—it never writes to your database—and provides a clear scorecard of findings.
This tool runs against your own already-indexed vector database and tells you: staleness, orphans, duplicates, and retrievable-after-delete—all read-only, so you can trust it won't touch your data.