Skip to main content

Selective disclosure: proof-unlocked documents

● Live on Zynapse

In one breath: you upload a document, a zero-knowledge proof is generated on the client side, and that proof becomes the key that unlocks a reading session over that exact data. You choose what gets disclosed. Nothing else exists, as far as the session is concerned.

The idea, with a filing cabinet​

Imagine a filing cabinet that doesn't have a lock. Instead, it has a rule: it will only open for someone who can prove they already hold the document inside, without showing the document to anyone on the way. And when it opens, it doesn't hand over the whole drawer. It answers exactly the question asked, and nothing more.

That's the mechanism:

  1. Client-side proof. Your document never leaves in the clear. A zero-knowledge proof of it is generated where the document lives: on your machine.
  2. The proof is the key. That proof file, not a password, not a cookie, is what opens a reading session bound to that specific data. No proof, no session. Wrong document, no session.
  3. Selective disclosure. Within the session you ask questions, and the session answers from the proven document. What's disclosed is what you ask for. The raw document is never on the table.

Two calls, that's the whole API​

# 1. generate the proof from your document (the key)
curl -s -X POST "https://zynapse.zkagi.ai/v1/generate_proof" \
  -H "X-API-Key: <YOUR_KEY>" \
  -F "file=@contract.pdf" -o proof.json

# 2. use the proof to unlock a Q&A session on that data
curl -s -X POST "https://zynapse.zkagi.ai/v1/ask" \
  -H "X-API-Key: <YOUR_KEY>" \
  -F "query=What are the termination clauses?" \
  -F "proof_file=@proof.json;filename=proof.json;type=application/json"

Full parameters, error codes and SDK snippets are in the Zynapse API reference.

Now couple it with the Sovereign tier​

Selective disclosure controls what is revealed. The Sovereign tier controls who could possibly see it while computing. Put them together and you get something genuinely rare: private introspection over documents.

The proof unlocks the session, and the session's inference itself runs as MPC: split across operators who each hold meaningless fragments, with the dealer role runnable on your own device. So you can interrogate a portfolio of contracts, a clinical file, a diligence data room, and at no point does any single machine, operator, or company (including us) hold both the question and the document in readable form. The answer arrives with a receipt proving how it was computed.

Ask your hardest questions of your most sensitive documents, and the infrastructure is built so that nobody, us included, gets to listen in.

Seen in the wild​

Our heart-health case study uses the same selective-disclosure principle in miniature: a Groth16 proof shows a private heart-risk score meets a policy threshold, and the verifier learns a YES/NO and nothing else. Same philosophy, clinical setting.

Where this stands

Proof-unlocked document Q&A (/v1/generate_proof + /v1/ask) is live on Zynapse today. Running that session end-to-end under the Sovereign MPC tier is active R&D and deployed per engagement; the coupling described above is the architecture we build toward, and what applies to your deployment is agreed in writing. See the Status board.