What does an Inferstamp proof actually prove?

That a specific answer follows from a specific set of model weights and a specific input, under the exact arithmetic of the circuit. The weights and the input stay private. Only short fingerprints of them appear in the receipt.

What does it not prove?

It does not prove the answer is correct, fair or useful, that the input describes reality, or that the provider did nothing else besides the proved computation. It also says nothing about how the input was collected or preprocessed.

Does checking a proof require access to the model?

No. Verification needs only the public receipt, the agreed fingerprints and a public verification key. Model weights are never sent to the verifier, which is why a provider can prove a result without disclosing its model.

How long does verification take?

Checking is fast and roughly constant: tens of milliseconds in a browser for the small circuit on this site, measured on 17 September 2026. Producing a proof is the expensive side, and that cost grows with model size.

Does this work for large language models?

Not today. The working prototype proves a small fixed-point linear model. Inferstamp does not currently support full LLM inference. Assessments focus on small numeric models and their verification requirements.

Every result
needs a receipt.

Your business depends on a model’s output. What backs it up? Inferstamp uses zero-knowledge proofs to connect a result to the agreed model and input, without sharing the model’s weights.

Try the proof How it works ↗

The result. The model. The proof.

Agree on the model

Agree on cryptographic fingerprints for the model and input. These identify what the proof must cover while keeping the underlying values private.

Attach the evidence

A proof accompanies the result. It shows that the output follows from the agreed model and input under the specified computation.

Verify independently

Check the receipt against the agreed fingerprints. You can verify the computation without access to the model’s weights or private input values.

Change the answer. Watch the proof fail.

Verify a sample result, then change its value and check again. This demo uses a small numeric model and a real proof. Verification runs in your browser.

The claim

Test data

Example Claimed answer

Type any number here, then run the check again.

The agreed fingerprints

Model: Loading…
Input: Loading…
Recorded: Loading…

Also change the proof’s claimed output Test whether the original proof can support your edited answer.

The check

● Runs on your device

Loading the example

Fetching the public files. No private data is needed.

Nothing is sent to a server

This checks a result recorded earlier. It proves the mathematics, but does not establish when the provider ran the model. Why that distinction matters ↗

Look inside the receipt

Inspect the result, model and input fingerprints, request details, and cryptographic proof. The model weights and input values are not included.

Download this receipt ↓

Evidence your customers can check.

For workflows where a model’s output needs independent verification.
All use cases ↗

Verify the agreed model

Give customers evidence that a result matches the model version they agreed to use.

Support accountable decisions

Keep verifiable evidence of the computation behind a score, price or eligibility result.

Protect your model’s IP

Let customers check a result without handing over the proprietary weights behind it.

Verify model updates

Check that a new checkpoint follows from the previous model and the agreed update rule.

The idea in 56 secondsSOUND + CAPTIONS

Know exactly what you can verify.

An Inferstamp proof verifies a specific computation. Model accuracy, data quality and actions outside that computation need their own checks.

Try the working demonstration for a small numeric model today. For your own workflow, start with a feasibility assessment. Full language-model inference is outside the current implementation.

Explore the technology and its scope ↗

Make your next model
decision an informed one.

Bring us a model and a verification problem. Leave with a clear recommendation, a defined proof scope and an estimate for a prototype.

€500 fixed-scope assessment

A working session on one model and one decision
A written statement of what can and cannot be proved
A comparison with signed receipts, re-running the model and hardware attestation
A prototype scope, cost estimate and key assumptions

The assessment covers feasibility and planning. Implementation is scoped separately. If you continue with a prototype, we credit the €500 assessment fee toward that work.

Every resultneeds a receipt.