Jev directory
CommunityJev Integrations & Adapters
@typia/jev
npm package that converts typia.llm.evaluation() questions into Jev's wire format, the format used by TypeSafe's API and SDK and by OpenRouter's Decisions API. Exposes toJevQuestions() and works with the TypeSafe client and the Vercel AI SDK.
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- Converts typia.llm.evaluation() questions to Jev's wire format through toJevQuestions().
- Jev spells the neutral yes/no question "boolean" as "noul"; choice and score questions are identical.
- decode() accepts Jev's native answers as they are, with no conversion step.
- Install with npm install @typia/jev typia @typesafe-ai/sdk, plus ttsc and typescript@rc as dev dependencies.
- MIT licensed and maintained inside the samchon/typia project.
Questions about this entry
Short answers drawn from the linked project or page.
Run npm install @typia/jev typia @typesafe-ai/sdk for the runtime packages, then npm install -D ttsc typescript@rc for the build-time dependencies the guide lists.
It converts the questions produced by typia.llm.evaluation() into Jev's wire format, the format used by TypeSafe's API and SDK and by OpenRouter's Decisions API, so they can be sent to a System One endpoint.
Neutral questions spell the yes/no question as "boolean", while Jev spells it as "noul". Choice and score questions are identical between the two formats.
No. decode() accepts Jev's native answers as they are, so the answers returned by the client can be passed straight into the evaluation's decode() method.
Jev's answers come rounded to two decimals in practice, which is the default for decimals, so decode() accepts them unchanged. For an endpoint with another precision, pass typia.llm.evaluation() instead.
Yes. Through the Vercel AI SDK you pass evaluation.questions to experimental_evaluate() as they are, and its TypeSafe and OpenRouter providers perform the conversion.
It is MIT licensed, matching the license of the samchon/typia project it ships alongside.
More Jev resources like this
npm adapter that adds optional TypeSafe Jev system one decisions to AgentRun DSL workflows. createJevRunner turns judge, pick, sift and route steps into typed Choice, Noul and Score answers, with retries, timeouts and OpenRouter support.
Jev Integrations & Adapters#typescript#sdk#agents
Community
The Vercel AI SDK provider for OpenRouter, including openrouter.decisionModel for typesafe/jev-1.13 evaluations with providerMetadata answers.
Jev Integrations & Adapters#community#typescript#openrouter
NestJS module that adds TypeSafe AI decisions to Loopstack workflows. It asks yes/no, choice, and score questions about workflow or agent state and returns probabilities, confidence values, and a legend per answer.
Jev Integrations & Adapters#typescript#sdk#classification
Community
npm package providing the TypeSafe Jev fast path for the MLola Browser Runtime: one request per decision cycle carrying the operation head plus speculative target heads, with responses validated before anything executes. Also ships planner and text adapters for OpenAI-compatible endpoints.
Jev Integrations & Adapters#typescript#browser#agents
Community
npm package that decides what happens to each new fact in an agent memory store: add, update, supersede, or skip. It calls TypeSafe Jev for typed answers with probabilities and ships as part of Oort, open-source memory for AI agents. Apache-2.0.
Jev Integrations & Adapters#typescript#agents#sdk
Community
A pm-cli extension and library that adds typed Jev decisions to project-management items. Commands cover triage, link, dedupe, ask, explain, gate, models, and doctor, with ranking, thresholds, and retries handled in code.
Jev Integrations & Adapters#typescript#cli#local
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.