API · Last updated 2026-09-21

Jev API replay endpoint

Replay means you kept state, questions, and the model pin, then you call the evaluation endpoint again. There is no documented /v1/replay.

This page is for builders who searched an extra “endpoint” name while wiring Jev / System One. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. We do not sell or issue keys. Confirm first-party routes on docs.typesafe.ai/api.

The public API reference documents one evaluation route — POST https://api.typesafe.ai/v1/systemone — plus GET /v1/models. This title is not a separate first-party URL in that reference. Treat the capability as something you implement around the official call.

Steps

1. Persist a replayable decision trace.
2. POST the same body (or the same pin + edited criteria for a counterfactual).
3. Compare policy actions, not only argmax.
4. Pin the id — jev-latest is not a replay key.

Map onto the official POST

Ordinary evaluation request/response: model, state, questions in; model, answers, usage out. Differences are yours to diff. Choice/Score include probabilities + confidence; Noul includes noul only. Question ids are echoed as keys; they were not sent to the model.

On the snapshot we read, jev-latest and jev-preview resolve to jev-1.13.0. Confirm before you treat two traces as the same weights. Typed slots are closed; values can still move across versions. Expecting bit-identical floats forever without a pin is the usual disappointment.

Each replay spends input tokens (vendor claim: $0.042 / million; output free). Official SDKs retry 429/529; a replay storm can 429 against the vendor-claimed 1200 RPM. Documented errors otherwise: 401, 422. Schema-safe equality of two choice strings is not proof the world agreed.

When not to replay

Do not replay production PII into an open Slack. Do not use replay as a TypeSafe-hosted job you GET later — that is your database. Promotion via replay is an environment process, not a path on api.typesafe.ai.

Pitfalls

• Expecting bit-identical floats forever without a pin.
• Replaying production PII into an open Slack.
• Calling /v1/replay.
• Diffing only argmax and missing a confidence collapse.

See also

• API hub — the POST you replay
• Decision trace — what to persist
• Environment promotion — controlled pin upgrades
• Offline evaluation — labeled fixtures

FAQ

Is /v1/replay documented?

No.

Deterministic outputs?

Typed slots are closed; values can still move across versions — deterministic outputs.

Promote via replay?

Environment promotion.

Is replay the same as idempotency?

No. Replay is for eval. Idempotency is for side effects you must not double.

What this page does not claim

• Official TypeSafe status, logos, or support.
• That jev.pro sells, issues, or proxies API keys.
• Extra first-party HTTP routes beyond POST /v1/systemone and GET /v1/models unless the API reference lists them.
• Invented latency SLOs, search volume, or unofficial prices. List price and rate limits are vendor claims — confirm on models.
• That a schema-constrained answer is automatically factually correct.

Disclaimer

This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.