Simple Jev: Open-Source Decision and Classification Engine Using Model Logits

LinkFeatherless.aiTool

Featherless.ai has introduced Simple Jev, an open-source implementation that turns open language models into structured decision engines. Instead of requiring a separate classifier head, it reads next-token logits for allowed labels and returns probabilities and choices directly as JSON.

Key points

  1. Uses open language and vision models (such as Gemma and Qwen3.6-35B-A3B-classifier) as classifiers without dedicated classifier heads.
  2. Operates by scoring next-token logits for specified criteria while reusing a shared prompt cache across questions.
  3. Accepts message context, full chat history, structured state, and images (vision supported for Gemma and Qwen).
  4. Returns structured JSON containing the chosen option, confidence score, and raw probability distribution.
  5. Offers 'RFDT' (Really Fancy Decision Training) to fine-tune decision logits using custom datasets or teacher models.
  6. Public demo API available without an API key (limited to 2k context and 2 RPS); production pricing on Featherless starts at $0.03 per million input tokens during beta.

Why it matters

Software applications and AI agents frequently require reliable categorization and routing rather than conversational responses. By evaluating token logits directly, developers gain deterministic, structured decisions with probability scores, avoiding fragile text parsing and reducing token consumption.

Analysis

Simple Jev targets a common inefficiency in LLM-assisted systems: relying on freeform text generation for discrete classification tasks. By leveraging next-token logits against constrained label sets and reusing prompt caches, the framework delivers fast, predictable outputs with calibrated confidence values. Its availability as open source alongside fine-tuning workflows (RFDT) provides an effective blueprint for high-throughput decision-making pipelines.

Mentioned links

Found in

Summary generated by AI (Gemini 3.8 Flash, 19 October 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Repository:NandhaKishorM/laya

    Laya: Non-Autoregressive Decision Engine for Fast In-Text Choices

    Laya is an open-source Python decision engine designed to deliver typed decisions—such as categorical choices, scores, or yes/no answers—across more than 100 languages. Instead of relying on autoregressive text generation, it produces decisions in a single forward pass taking approximately 33 ms.

    26kstarsPython

    AI & LLMs· Tool

  • X post:Ruben Hassid

    TypeSafe's Jev: Ultra-Fast AI Classification Without Chatbot Latency

    Ruben Hassid highlights Jev by TypeSafe as a fast, low-cost decision engine designed to bypass standard LLM text generation. Rather than generating word-by-word paragraphs, Jev answers predefined multiple-choice questions with a direct decision and confidence score. Hassid details how to integrate it into Claude Code and Codex for tasks like bulk contact classification.

    1.6kbookmarks112.5kviews

    AI & LLMs· Tool

  • Video

    Video:CURT

    Jev Design Patterns: Operating Workflows with Decision Models

    This video introduces Jev by TypeSafe, a specialised decision model designed for low-cost, low-latency micro-decisions rather than text generation. It covers three decision types, ten functional shapes, and four architectural design patterns through practical examples.

    AI & LLMs· Tutorial

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.