Why does noul carry no confidence when choice and score do?

No confidence is reported for noul because the probability already expresses certainty. Choice and score report a distribution-derived confidence you threshold to decide whether to act or escalate.

Will constructing the client hit the network or fail on a bad key?

No. Construction performs no I/O and cannot fail on network, key, or rate-limit problems. Discovery is lazy per client, and an explicit warm method reports failure as false instead of throwing.

How do retries separate rate limits from genuine server outages?

Yes. Rate limiting is a separate exception from server failures, carries the Retry-After value, and defaults match the official SDKs with two retries, doubling delay, and a ceiling.

Do I need to pin a model name or will the default go stale?

No. The default is jev-latest, a server-resolved alias for the current flagship, so callers get a sensible model without naming one. The model set is open, so no enum is provided.

How do I install Jev.Sdk if it is not on nuget.org?

Neither package is published to nuget.org. The build workflow produces the .nupkg files as downloadable artifacts on every push to main, so you pull them from the CI artifacts rather than installing from nuget.

How does Jev.Sdk model discovery and its caching work?

GET /v1/models is called lazily and cached per client for an hour by default, so a shared cache cannot leak one account's model list to another. Concurrent first reads share a single request, and ModelCacheDuration can shorten or disable the cache.

Does Jev.Sdk validate my request before sending it?

Yes. Local validation runs before any network call and reports every problem at once, so one round of feedback catches all mistakes. A Score question with a single level is rejected locally even though the API accepts it, because a one-level scale returns a constant.

CommunityJev Repos & SDKs

Jev.Sdk

Independent .NET client for the System One API with typed noul, choice, and score questions, DependencyInjection integration, cached model discovery, local validation, and typed errors.

See highlights

Open github.com

Stars: 1
Language: C#
License: MIT
Published by: Community
Author: nothingmn
Last verified: Oct 2026

nothingmn/Jev.Sdk

User repository on GitHub

View on GitHub

A .NET client library for the [TypeSafe AI](https://typesafe.ai) System One API (`https://api.typesafe.ai`).

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A .NET client library for the TypeSafe AI System One API (`. Ask typed questions about a body of content and get structured, probability-backed answers back. Every question is declared with a kind and answered within the constraints you supplied, so nothing has to be parsed out of prose.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: C# / .NET

Highlights

Noul answers with a probability and no confidence; choice and score add distribution-derived confidence for gating.
Many questions run in one request against shared state, evaluated independently in parallel, with JSON structured state.
Model discovery uses GET /v1/models with a per-client hourly cache, no I/O at construction, and shared concurrent reads.
Every failure derives from JevException with request ID and endpoint; rate limits stay distinct from server failures.

Watch out

Independent library with no TypeSafe affiliation; packages ship as workflow artifacts on pushes to main rather than nuget.org. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

FYIsoft.Extensions.AI.Providers

MIT-licensed Microsoft.Extensions.AI providers for Anthropic Claude and TypeSafe Jev, plus a SupportDesk console sample wiring triage, drafting, and review together.