access-regression gate for enterprise AI · runs locally · exit 0 or 1

Permissions don't travel
with your data.

When documents are copied into a RAG or agent index, their access controls get left behind. People change teams, sharing gets revoked, an embedding model gets upgraded — and the index quietly keeps serving what it learned on day one. raggate proves, before every rollout, that no identity can retrieve chunks it isn't authorized to see — and keeps replayable evidence after.

Why your permission system can't see it

Nothing changed in your authz. Everything changed in reach.

Authorization engines answer "can X read Y?" — and keep answering correctly while retrieval quietly changes which Y come back. Three regressions we plant in our benchmark corpus, and catch on every run:

revoked · still indexed

The employee who left the project

Bob loses project access on Monday. The project's chunks were embedded last month. His queries still surface them — the index never heard about the revocation.

gate result: 0 → 6 leaks the moment permissions change
cross-tenant · similarity trap

The two customers with identical contracts

Tenant A's pricing query is a near-perfect match for tenant B's pricing template. Pure vector similarity doesn't know what a tenant is.

gate result: 6 leaks, both directions, pre-existing
pipeline change · silent regression

The embedding upgrade

A "better" embedding model reaches further semantically. Permissions untouched — and documents that were unreachable yesterday are in answers today.

gate result: 10 forbidden docs newly reachable, ACLs unchanged
What you get

A verdict your security review can act on. Evidence your audit can replay.

raggate run --db out/gate.db --snapshot acl/v2.yaml --scenarios scenarios/
gate run against snapshot v2 (final_k=5, tau=0.3)
  S1: 6 scenarios, 6 leaked · revoked access, stale index
  S2: 6 scenarios, 6 leaked · cross-tenant recall
  S3: 4 scenarios, 4 leaked · public-then-private doc
  NEG: 10 scenarios, 0 leaked · zero false positives
  LEAK [S1] bob@acme → acme-atlas-01
evidence: 260 records → out/evidence.jsonl
VERDICT: FAIL — access regression detected (exit 1 · blocks the pipeline)
Regression testing, not probing
Deterministic replay of scenario queries as real identities against your ACL ground truth — same inputs, same verdict, every run. Not LLM-judged red-teaming.
Replay-complete evidence
Every retrieved chunk becomes one JSONL record — who asked, what came back, the authorization decision, the policy version. Re-derive any verdict months later without re-running retrieval. IDs and decisions only; never document content.
Made for CI
Exit 1 on any leak. Gate embedding upgrades, chunking changes, re-ranker swaps, ACL syncs — every change to the pipeline, not just the first launch.
A report you can put in front of anyone
Single-file HTML: verdict banner, per-scenario evidence, access-surface diff. Opens from file://, no server, nothing leaves your machine.
How to try

Bring your own snapshot. Run it where your data lives.

Export a permission snapshot

A point-in-time YAML of who-can-read-what, pulled from your source of truth — Drive, SharePoint, Lark, GitLab, your IdP, or your authz service. Open spec, one file, raggate validate-snapshot checks it before anything runs.

Point the gate at your index

Ingest your corpus (or adapt your retriever behind one interface: retrieve(query, identity, top_k)). Write scenarios: identities, queries, and what they must never see.

Get the verdict — and keep the evidence

raggate run exits 0 or 1. Wire it into CI so every pipeline change re-proves the access boundary. Evidence and reports accumulate as your audit trail.

Local-first, by design. The CLI runs entirely in your environment. Your documents, your ACLs, and your identities never leave it — a product that verifies leak-prevention shouldn't start by asking you to leak. Only redacted evidence summaries are ever optional to share.
Pricing

Priced per protected pipeline. Test as often as you like.

A pipeline is one retrieval index / knowledge base / agent app — its own corpus, retriever config, permission snapshots, and scenarios. Environments don't count extra. Runs are unlimited: a gate should never charge you for testing more.

Design Partner Pilot
$2.5k–$7.5k
60–90 days · fixed scope · start here
  • Prove zero-leak rollout against your real ACLs, embeddings & retrieval stack
  • Your scenarios + our planted-leak suite
  • Full per-query evidence + audit-ready report
  • Runs entirely in your environment
Request a pilot
Production
from $750
per month · after your pilot
  • CI gate on every change
  • Access-surface diff across configs
  • Report history & run archive
  • Local license, air-gapped OK
  • Priced by pipeline / repo / corpus / environment
Talk to us
Enterprise
Custom
annual · unlimited scale
  • SSO, private deployment
  • Audit package & evidence retention
  • Permission-source connectors
  • Dedicated support
Talk to us

Just want to try the CLI? — Developer Sandbox, $249/mo

One pipeline, non-production, community support. The full gate and evidence reports, for kicking the tires. Most teams start with the pilot instead — it answers the question your security review actually asks.

Get the Sandbox

A pipeline = one retrieval index / knowledge base / agent app. Environments don't count extra; runs are unlimited — a gate should never charge you for testing more. Today every button reaches a human who responds fast.

Get started

Book a pilot — or ask us anything.

Tell us what you're shipping and who blocks you at launch. We reply from a human, fast. Nothing here touches your corpus or ACLs.

A human reads every one — usually same day. Prefer email? [email protected]