Jev Tutorial
English
English 简体中文
Submit a project
REPOSITORY ANALYSIS

ai

vercel · SDK & Decision Frameworks

The TypeSafe provider in AI SDK lets TypeScript applications call Jev through the shared evaluate interface.

E2 · source reviewedPublic material or source evidence was reviewed. A fixed-version source review requires checking a pinned revision; this is not a reproduction, performance endorsement, or safety guarantee.
Open GitHub repository ↗ Back to project radar
Stars
26,922
Forks
5,180
License
—
Last commit
2026-09-23
JEV · DECISION POINT

What Jev does here

The TypeSafe evaluation provider maps AI SDK boolean questions to native Noul and converts the responses back. Native Choice/Score confidence is retained in providerMetadata.typesafe.confidence rather than the generic answer object.

WHY IT HELPS

Allows a TypeScript AI SDK application to use a common evaluation interface while keeping text generation and finite decisions separate.

SOURCE NOTE

Pinned adapter source reviewed. No live inference was run; native and AI SDK response fields differ, so this integration is not evidence that a generic evaluation probability equals TypeSafe confidence.

REVIEWED SOURCE LINKS
Source 1 ↗Source 2 ↗Source 3 ↗
EDITORIAL ANALYSIS · TASK AND IMPLEMENTATION

What task does this project fit?

Use Jev through the AI SDK evaluation interface instead of pretending it is a text-generation model.

Material analysed：

From input to action

  1. Supply choice, score or boolean evaluation questions.
  2. Map boolean requests to native noul and call TypeSafe.
  3. Map native answers back and expose additional confidence in provider metadata.

Implementation details that matter

Boolean is a provider-boundary translation

The adapter turns evaluation boolean questions into noul, then translates answer.noul back to probability. Application code must use the AI SDK shape, not native SDK property names.

Inspect the supporting source ↗

Confidence is preserved outside generic answers

Native Choice/Score confidence is collected in providerMetadata.typesafe.confidence. Generic answer probabilities remain a separate field and must not silently replace that statistic.

Inspect the supporting source ↗

When to choose it

Choose this when an AI SDK application needs evaluate-style decisions and a common provider interface. Keep generateText for tasks that actually require generated prose.

How to inspect it yourself

Read question mapping near the POST body and response mapping together. With a fake response, check boolean probability and per-question confidence metadata separately.

Limits of this review

  • No provider inference was run here.
  • This fixed adapter has its own option/level validation and rounding metadata; confirm the installed package version.
Primary material used in this analysis