RAG governance — retrieval & provenance authority demo

A retrieval-augmented AI assistant answers a question from retrieved context. KYE Protocol™ checks the consequential part: was the retrieval authorised, do the answer's citations match the sources actually retrieved (no fabricated cite), were those sources licensed for this use, and is the answer replayable from pinned provenance? It then seals a signed, replay-provable Evidence Pack™. This is RAG authority and provenance — not a retrieval engine.

All logic runs in your browser. The Evidence Pack™ hash is a real SHA-256 (Secure Hash Algorithm 256-bit) over canonical JSON, recomputable in your terminal.

Start the demo → License authority demo
1 — The retrieval + the answer

Choose what the assistant did

You own the knowledge base. The assistant retrieved context, then drafted a cited answer. Pick the conditions below — an out-of-scope corpus, a fabricated citation, or an unlicensed training use — then run the admissibility check.

2 — Action-admissibility gate

Five rules, in order

The gate runs five rules. Each returns ALLOW or REFUSE. A single refusal stops the answer and shows the exact reason code you receive — the same provenance discipline the EU AI Act puts on the training data behind general-purpose AI models.

    3 — RAG provenance Evidence Pack

    The artefact you show an auditor or a court

    When every rule allows the answer, KYE seals the Evidence Pack™. It binds the question, the authorised retrieval, the pinned sources, the citation set, and the rule results — verifiable offline from published keys alone, valid at T=0. If the answer is later disputed, the pack proves it was grounded in sources the assistant was authorised to use.

    Run the check to build the Evidence Pack™.

    
      
    
    
    

    
    
      Honest scope
      
    What this demo proves — and what it does not
    
      
    KYE governs the authority and provenance of a RAG answer — was the retrieval authorised, are the citations pinned to sources actually retrieved, were those sources licensed for this use, and is the answer replayable. KYE does not store your vectors, run embeddings, or orchestrate retrieval — that is your retrieval stack, and KYE does not replace it. It reuses the canonical source-pin primitive (kye.evidence.tool_call_pin.v1) and the §62 "no claim without a cited, pinned source" discipline. The retrieval quality and the answer's truth remain yours.