Simple Jev
Simple Jev (github.
Overview
Simple Jev (github.com/featherless-ai/simple-jev) from Featherless AI turns compatible open models into structured decision endpoints. Clients send shared state plus typed questions; the server reads next-token logits and assembles JSON for Choice, Score, and Noul answers. The model never free-writes JSON. A public demo API at simple-jev-demo-api.featherless.ai allows two requests per second with a two thousand token context cap and no API key; production paths run on Featherless plans or your own HF server from the repo.
Problem: Teams want Jev-shaped classifier HTTP without training a separate head or parsing model-generated JSON blobs.
Built for: ML engineers standardizing on Hugging Face weights who need a /v1/classifier surface for agents and eval scripts.
First indexed on Jev Directory: 2026-09-22
Creator and team
- Name
- Featherless AI
- Organization
- Featherless AI
- Handle
- @featherless-ai
How Jev is used
- Role in the product flow
- Hosted or self-hosted classifier API built from HF model logits
- Primitives
- ChoiceScoreNoul
- State in
- Shared context string plus questions map with type choice, score, or noul criteria per OpenAPI examples in the repo.
- Decision out
- Structured JSON response with probabilities per option; assembled server-side from logits.
- Validate request against shared common/ schemas
- Run one forward pass over state and question prompts
- Score each criterion token from logits
- Return versioned JSON without decoding a completion
Simple Jev is the bring-your-own-weights cousin of hosted classifier.dev. Agents keep the same mental model: pack state once, fan out questions, threshold probabilities in code. Featherless hosts a playground and demo API so you can curl a Gemma classifier ID before you sync weights locally. The common/ Python package holds validation and scoring rules so alternate inference backends can share behavior. Mention OpenRouter or TypeSafe hosted routes only when you compare latency; this repo is for open HF pipelines.
Sourced performance claims
- Public GitHub repo featherless-ai/simple-jev had about four hundred eighty four stars when this listing was drafted.Source: GitHub star count September 2026
- Demo API documents 2k token context and 2 RPS limits without authentication.Source: github.com/featherless-ai/simple-jev README
Features and stack
Features
- POST /v1/classifier compatible HTTP surface
- Public no-login demo API with documented rate limits
- Playground and docs at simple-jev.featherless.ai
- Self-host path with Hugging Face Transformers server
Stack
- Python
- PyTorch
- Hugging Face Transformers
- Featherless inference
Pricing: Demo tier is free with tight limits; Featherless paid plans raise caps per featherless.ai pricing.
Links
FAQ
- Does the model emit JSON?
- No. README states the server constructs JSON from logits; the model does not complete a JSON string.
- Which models work?
- GET /v1/models on the demo API lists served classifier IDs such as featherless-ai/gemma-4-26B-A4B-classifier; self-host docs cover adding weights.
- Is this TypeSafe hosted Jev?
- No. Simple Jev runs open models on Featherless or your hardware. Use classifier.dev or TypeSafe keys when you want the closed jev model.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- System One model
The model family behind Jev: parallel typed questions, one forward pass, probabilities you can threshold.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.