WaterSheep

Apache-2.0 ModernBERT decision model answering Noul, Choice, Score and multi-label questions, with a local Jev-compatible server.

ServicebetaApache-2.0

View on GitHub ↗Live demo ↗

Overview

WaterSheep is an open-weight decision model for text that answers yes/no, single-choice, rating and multi-label questions with a probability for every option. It loads as a Hugging Face transformers pipeline and can run offline after download, or in the browser via ONNX according to the list.

For Jev users, it runs as a local server that answers POST /v1/systemone requests, so TypeSafe's Python SDK works unchanged after setting TYPESAFE_BASE_URL; any API key value works locally, and multi-label questions are supported as plain JSON. It can also be deployed as a Hugging Face Inference Endpoint. The list reports an expected calibration error of 0.026 on its test split. It is Apache-2.0 and independent of TypeSafe, with a demo Space and video.

What the decision model decides

Type	Question	Answers
Choice	Which team should handle this?	billing · shipping · support
Yes / no	Yes/no question about the text	yes · no
Other	Multi-label: which of these labels apply?	—

How it works

Install transformers and load the pipeline, or run watersheep --serve.
Point TypeSafe's SDK at the local base URL if desired.
Ask Noul, Choice, Score or multi-label questions.
Use per-option probabilities.

Strengths

Drop-in local replacement for Jev's API.
Adds multi-label questions.
Reported ECE of 0.026.

Caveats

Text only.
Accuracy relative to Jev not stated in the README excerpt.

Getting started

pip install git+https://github.com/SamratDuttaOfficial/WaterSheep && watersheep --model samratduttaofficial/WaterSheep --serve

At a glance

Type: Service
Status: beta
Language: Python
Runtime: transformers, local HTTP server, ONNX
Has code: yes
License: Apache-2.0

Use cases and integrations

Calibration & Research264

How curated lists describe it

Open alternative: an Apache-2.0 ModernBERT fine-tune that answers `Noul`, `Choice`, `Score` and multi-label questions with a probability for every option, serves Jev's `POST /v1/systemone` locally so TypeSafe's Python SDK runs against it unchanged, and reports 0.026 expected calibration error on its test split.
yibie/awesome-jev
Apache-2.0 open-weight decision model for text: Noul, Choice, Score, and multi-label questions get a calibrated probability for every option, served by a local `/v1/systemone` server that TypeSafe's Python SDK works with unchanged
AnotiaWang/awesome-decision-models
Open-weight Jev alternative that serves /v1/systemone locally, with calibrated Noul, Choice, Score and multi-label answers.
logicrw/awesome-jev-projects
A decision model that returns calibrated answers to yes/no, choice, score and…
wh000wh000/awesome-jev-live
Independent open-weight Jev alternative (Apache-2.0) behind a local server that accepts Jev's `POST /v1/systemone` request format; it answers Noul, Choice, Score, and multi-label questions about text with a probability for every option, and TypeSafe's Python SDK works against it unchanged.
cobanov/awesome-jev
Open, Apache-2.0 alternative to Jev that runs locally. Its server accepts Jev's /v1/systemone requests, so TypeSafe's Python SDK works against it, and it adds multi-label questions.
hellogumbo/awesome-jev
An open-weight model that answers Choice, Score and Noul questions, plus multi-label ones, with a probability for every option. Serves a local `POST /v1/systemone` endpoint; its README says TypeSafe's Python SDK works against it by changing the base URL. Also runs in the browser through ONNX.
MrJev/awesome-jev

Frequently asked questions

Does TypeSafe's SDK work with it?

Yes, set TYPESAFE_BASE_URL to the local server.

What extra question type does it support?

Multi-label questions.

Do I need a key?

No; any key value works locally.

Related projects

JevBERT
Unofficial proof of concept serving Jev's /v1/systemone shape from a local BERT-style zero-shot classifier, with honest caveats.
ServiceprototypePython
gojev
Go harness for System One models: hosted Jev, open Kev in-process via llama.cpp, and Cloudflare Clef, with typed helpers and a CLI.
ServicebetaGo
jev
TypeScript library for one-line local decisions with open models (kev-0.6b, kev-4b, open-jev), no API keys or cloud.
ProjectbetaTypeScript
jev-foundation-models
Swift 6 bridge that runs System One decision models behind Apple's Foundation Models API, on device with Laya or via hosted endpoints.
ProjectbetaSwift
laya-system-one
Zero-dependency npm package that runs Laya natively or as WebAssembly in Node, Bun and browsers, drop-in for Jev's /v1/systemone.
ProjectbetaJavaScript
OmniStudio
Local-first desktop AI workbench for llama.cpp, vLLM and SGLang that includes a JEV typed-decision page and /v1/systemone gateway.
Servicebeta