ElBruno.AI.Decisions.Jev 0.6.2

ElBruno.AI.Decisions.Jev

A community .NET 10 client for the official TypeSafe AI Jev service. It is one provider of the ElBruno.AI.Decisions family, which also includes ElBruno.AI.Decisions (provider-neutral IDecisionClient), ElBruno.AI.Decisions.Foundry (Microsoft-Decision-1) and ElBruno.AI.Decisions.Ollama (local models).

Tentative early-access release (0.6.2). Live Jev and local Ollama compatibility remains unverified, and live testing is deferred. Offline test and package validation success does not prove real-service behavior. Evaluate before production use; APIs may change before 1.0. NuGet classifies 0.6.2 as stable because it has no prerelease suffix, but this package is still explicitly tentative.

dotnet add package ElBruno.AI.Decisions.Jev --version 0.6.2

Capabilities

The Foundry package uses /providers/microsoft/v1/systemone and the named TypeSafe-style question contract documented by Microsoft. Choice, Score and Assess were smoke-tested live with API-key authentication on October 9, 2026. This does not establish production readiness or probability calibration.

The current-source Ollama provider uses /v1/systemone with local decision models such as nimble (Ollama v0.35.0+), not chat models. Its native System One correction requires 0.6.1 or later; 0.6.0 contains the old chat-logprob path. It preserves native confidence (distribution concentration) separately from label probability. The shared text API exposes one Choice, Score (2-10 ordered levels), or Assess question per call; image and batch requests are not exposed. See the Foundry and Ollama response examples.

• Choice classification with full probability distributions and confidence.
• Score evaluation against ordered rubrics, preserving fractional results and legends.
• Noul proposition probability without hidden boolean thresholds.
• Structured state and criteria, typed question handles, model discovery and pinning.
• Explicit loopback-only local Laya inference with optional bearer authentication.
• Microsoft.Extensions.AI function tools, chat routing, and assessment composition.
• Async HTTP, cancellation, bounded responses, explicit errors and controlled retries.

Jev is not a generative chat or embedding service. This package does not invent image, speech, or realtime endpoints. It is not an official TypeSafe AI SDK and does not target the independent jevtypesafeai.com proxy.

Quickstart

using ElBruno.AI.Decisions.Jev;

using var client = new JevClient(new JevClientOptions
{
    ApiKey = configuration["Jev:ApiKey"]
        ?? throw new InvalidOperationException("Configure Jev:ApiKey."),
    DefaultModel = JevModels.Version1_13_0
});
var question = new JevQuestionKey<JevNoulAnswer>("relevant");
var request = new JevDecisionRequest("A .NET developer asks about dependency injection.")
    .WithQuestion(question, new JevNoulQuestion("Is this about .NET development?"));
var response = await client.EvaluateAsync(request, cancellationToken);
Console.WriteLine(response.GetAnswer(question).Probability);

The consuming application supplies configuration and cancellationToken. Use .NET user-secrets for development or your application's production secret provider. Do not hardcode API keys.

Microsoft.Extensions.AI integrations wrap your own chat clients; a Jev credential does not enable another provider. Review configured confidence thresholds and model versions before making consequential decisions. Assessments are not security guarantees.

Official service documentation: https://docs.typesafe.ai/

The maintainer has authorized this tentative release before live verification. See the project repository for runnable samples, configuration, tests, known limitations, and release instructions.