TinyZKP
Free evaluation path

Check fit before you move a proof job.

The Community engine owns compatibility, automatic mode planning, RAM estimation, and scratch-headroom checks. It rejects unsafe or unsupported jobs before expensive ingestion.

Run the free doctorReview the exact profile
Signed evaluation release available. Download and verify the exact Community doctor identity published in discovery.json before use.

Expected local flow

tinyzkp-engine doctor --job job.json

The signed evaluation release will emit one versioned DoctorReportV1 result. Keep that report local. Never send its contents, a witness, trace chunk, license key, entitlement, secret, private path, or proprietary AIR source. A founding-validation record may contain only the report digest and standardized compatibility result.

Good fit

Continue evaluating when

  • Your production target is Linux x86-64.
  • Your workload matches the exact published Plonky3 profile.
  • You have a measured OOM, high-memory machine cost, or restart-loss problem.
  • Local NVMe scratch under your control is acceptable.

Stop when

  • You need lookups, buses, multi-table AIRs, recursion, a GPU, or a custom field.
  • You need hosted proving, a service bureau, an SLA, custom AIR work, SSO, or private branches.
  • You cannot keep witness-derived scratch data on customer-managed storage.
  • The doctor reports an unsupported profile or unsafe path.

Local ROI worksheet

TinyZKP does not publish a universal ROI claim. This worksheet runs only in your browser, sends nothing, and stores nothing. Enter your own annual avoidable costs; its calculation is not benchmark or purchase evidence.

avoidable cost = capacity + (restart hours × hourly cost) + delay cost

Enter your documented costs to calculate a local comparison.
Decision rule: continue only when the documented annual avoidable cost materially exceeds the engineering time to adopt and operate the engine, and integration remains self-service. A hypothetical estimate is not benchmark evidence. The Guard subscription is withdrawn and is not sold at any price, so nothing on this page is a purchase decision.

What to do with a result

  1. If compatible, compare the estimate with your measured baseline and the published evidence contract.
  2. If incompatible, use the exact reason link in troubleshooting; do not work around it.
  3. If the report appears wrong, open a public issue with synthetic inputs and the exact release identity.