RAGproof.io — vector DB audit

Your RAG index is quietly rotting.

RAGproof audits your vector database read-only — pgvector, Qdrant, Weaviate, Chroma, or Pinecone — and reports exactly which content is stale, orphaned, duplicated, or deleted-but-still-retrievable. Nothing leaves your environment.

Used only to respond to this request.

audit_report.txt
// index health — acme-prod-index
31% of vectors older than source document
4,812 orphaned embeddings (no source record)
1,203 duplicate clusters, avg. 3.4 copies
12 deleted documents still retrievable
→ GDPR Art. 17 erasure exposure

01 — What you get

One report. Every stale, orphaned, and legally exposed vector, named.

  • 01

    Freshness map

    Every vector plotted against its source document's last-modified date — surfaces drift at a glance.

  • 02

    Orphan list

    Embeddings with no matching source record left in your system of record.

  • 03

    Duplicate clusters

    Near-identical chunks re-embedded across ingestion runs, grouped and counted.

  • 04

    Ghost-vector / erasure check

    Deleted-but-still-retrievable content — flagged against GDPR Article 17.

  • 05

    Prioritized fix plan

    Ranked remediation steps, ordered by retrieval-quality and compliance risk.

02 — How it works

Three steps. Nothing installed permanently, nothing exported.

  1. 01

    Read-only connection

    You grant read-only credentials scoped to the index. We never write, delete, or modify.

  2. 02

    Self-hosted scan, your VPC

    The scan runs inside your environment. No vectors or source content leave your infrastructure.

  3. 03

    Report + walkthrough call

    One-page report, delivered and walked through live — findings, risk, and fix order.

03 — Why now

Why now

Staleness is invisible — until users stop trusting the answers.

There's no error, no alert, no failed request. Retrieval keeps returning results — they're just increasingly wrong. By the time users notice, trust in the system is already gone.

Deleted-but-retrievable content is a GDPR Article 17 problem.

If a document is erased from your source of record but its embedding still surfaces in retrieval, the right to erasure hasn't actually been honored. That's a compliance finding, not an edge case.