Your AI finds everything. Rasepi decides what it can use.

Before your AI writes a response, Rasepi checks every source it finds. Is it up to date? Verified? Approved for this purpose? Only what passes muster makes it through.

🤖
The AI search finds 20 sources
Support bot · public customer channel
8 sources approved
Up-to-date, verified, approved for the channel
ALLOW
5 sources with a warning
Valid, but review expires soon · Citation required
WARN
7 sources blocked
Expired, source changed, or never approved
BLOCK
The AI only responds using verified sources. Every decision in the provenance ledger

It happens in every company that relies on an AI assistant for its own knowledge. A customer asks the support bot about API setup. The search finds twelve relevant documents. Three of them are outdated. One was never intended for customers. The AI cannot tell the difference. It constructs a fluid, convincing answer from everything.

Relevant is not the same as reliable. This is exactly where the Trust Gate comes in.

A verification step between searching and answering.

You don’t change anything about your AI. Just a single verification step before the answer is generated. Rasepi doesn’t read any prompts and doesn’t sit between you and your model.

1

A question comes in

A person, a chatbot, or an agent needs information. Your AI searches as usual and finds, say, 20 sources.

2

Rasepi checks every source

Timeliness, verification status, ownership, origin, approvals, and the context of the request: Who is asking, why, and through which channel? This results in a decision for each source: ALLOW, WARN, BLOCK, or ESCALATE. Always with a justification.

3

The AI responds only from verified sources

Blocked sources never receive the response. Warnings may trigger citation requirements or notifications. The response is based on knowledge you can trust.

4

The decision is logged

Which sources were verified, which were approved, which were blocked, and why. The provenance ledger records it all, for audits and troubleshooting.

Three integration points. One of them is already a perfect fit today.

For the technology behind it: The Trust Gate can be integrated at three points in your pipeline.

A

Before retrieval

Your application first asks: Which sources are even permitted for this use case? Strict, good for regulated environments.

B

After retrieval, before generation

The retriever finds candidates, Rasepi filters them. The easiest way to get started and our recommendation. A single API call in your existing pipeline.

C

After generation

Rasepi checks whether the cited sources were permissible. Useful as a second line of defense, weaker than the gating step before it.

The same document. Four answers.

Trust is not a property of a document. It is a decision made in context: Who is asking, for what purpose, and through which channel? A score alone cannot do that.

OAuth Migration Guide
Engineering · Review overdue for 12 days · linked API reference recently changed
👤 Employee reading internally
WARN
Review overdue, but the responsible owner is active.
🤖 Internal AI assistant
WARN
Allowed for internal AI, with citation required.
💬 Customer support bot
BLOCK
Linked API reference changed since last approval. Customer responses require a new approval.
Autonomous Agent
BLOCK
Document is not approved for active agents.

Keep your systems. Add trust.

The Trust Gate does not require migration. Rasepi connects to the systems where your knowledge resides today, catalogs the content, and calculates trust based on it. It doesn’t matter which systems they are.

  • Every source is cataloged: wikis, documentation platforms, Git repositories, ticket systems
  • Broad cataloging, deep analysis only where necessary. Even with very large collections
  • Trust status visible where work is done, via API and badges
  • No lock-in: Your knowledge stays where it is
Connected Knowledge 4 Source Systems
Company wiki
2,418 pages cataloged
61
142 reviews overdue 38 sources modified
Developer documentation
312 pages cataloged
88
Webhooks active Reviews current
Team notes
876 pages cataloged
34
No owner for 61% Blocked for AI

Every decision leaves a record.

Which sources were reviewed? Which were approved, which were blocked, and why? Which version was valid at the time of the response? The Provenance Ledger answers the question that arises after every incident and in every audit.

“Who or what relied on which knowledge, and why was that permitted?”

Blocked sources automatically trigger review requests to the owners. The Gate not only prevents damage, it drives the maintenance of your knowledge.

One endpoint, every pipeline.

The Trust Gate uses a simple HTTP API: sources in, decisions out. Machine authentication via API key or client credentials. Works with any RAG framework, any chatbot platform, and in CI/CD pipelines. All endpoints, request examples, and SDKs can be found in the API documentation.

Go to the API documentation →

Only feed your AI knowledge
you would trust yourself.

Trust Gate is part of the private beta. Sign up and tell us about your pipeline.

We invite teams in waves.