Can Gutsy run without a GPU or an API key?

Yes. It runs a quantized 0.8B GGUF on an ordinary CPU through llama.cpp, makes no network calls and has no per-call cost. On long states the README reports a median of 4.3 s and a 95th percentile of 27 s on JevBench's hard tier.

Is Gutsy compatible with existing Jev clients?

It follows the Jev-style Decisions format over /v1/systemone with choice, score and noul questions, so the README says existing clients mostly need only a new base URL pointing at the local server.

How good are the probabilities?

Validation calibration error is 0.022, so a 0.8 means roughly 80%, and 96 to 97% of answers are unchanged when the options are shuffled. It is also deterministic: the same request returns the same probabilities on the same machine and build.

How does it compare with Jev itself?

It scored 0.732 on JevBench public items against Jev's 0.866, with financial sentiment at 0.927 and RAGTruth at 0.813 in the project's own benchmarks. Banking77 is held out of training, so that comparison is zero-shot for all three systems, and there is no official JevBench result yet.

What are the documented weak spots?

Date and number arithmetic is near chance, so compute in code and ask about the result. It is weak on new multi-step rulebooks at 0.32 on JevBench long_policy, still hedges in about a third of cases the rules already settle, struggles on close pairs and world knowledge, and calibration degrades to 0.137 error on hard unfamiliar items against 0.022 on validation. Input is English only.

How do I install and run it?

Clone the repository, `pip install -e gutsy/gutsy-inference` with llama-cpp-python 0.3.35 or newer, download the GGUF with hf download, copy models.example.json to models.json, then run `gutsy-inference check` and `gutsy-inference serve`. The server listens on 127.0.0.1:8765 and the default model is the Q8_0 file; send gutsy-0.8b-v04-q4 for the smaller 505 MB one at 25 to 30% more speed.

How was it trained, and what are the terms?

A full fine-tune of Qwen3.5-0.8B on about 105,000 questions in one epoch on one 80 GB GPU, roughly an hour, with cross-entropy plus Brier loss on soft labels. A third of the data comes from an open teacher model, Qwen3.8-27B, kept only when two independent passes agreed. Code and weights are Apache 2.0, with no closed-model output anywhere in the training data and the base and teacher models also Apache 2.0.

CommunityJev Models & Reimplementations

Gutsy

An Apache-2.0 0.8B decision model that returns calibrated probabilities for yes/no, choice and score questions on CPU. Distributed as GGUF for llama.cpp and served behind a Jev-style /v1/systemone API, so existing Jev clients mostly need only a new base URL.

See highlights

Open github.com

Stars: 77
Language: HTML
License: APACHE-2.0
Published by: Community
Last verified: Oct 2026

kouhxp/gutsy

User repository on GitHub

View on GitHub

Local decision model with calibrated probabilities: send a state and yes/no, choice or score questions, get a probability for every option. 0.8B GGUF on CPU, Jev-style API.

Forks: 2
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Local decision models with calibrated probabilities. Send a state (text or JSON) and typed questions (yes/no, choice, score); get a probability for every option back. No text generation, no network calls, no per-call cost. The API follows the Jev-style Decisions format, so existing clients mostly need only a new base URL. Calibrated. A 0.8 means roughly 80% (validation calibration error 0.022). Local and private. Runs on CPU; your data never leaves the machine.

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

Tagscommunity gguf local-inference calibration

Works forJev for Classification Jev for Routing Jev for Scoring Jev for Moderation

Used inJev for Customer Support Jev for Finance & Fintech

Languages: Python
Platforms: CLI API Browser
Capabilities: Open sourceSelf-hostedRuns locally

Highlights

gutsy-0.8b v0.4 is a fine-tune of Qwen3.5-0.8B shipped as Q8_0 (775 MB) or Q4_K_M (505 MB) GGUF files on Hugging Face.
Runs on an ordinary CPU through llama.cpp with no network calls and no per-call cost; the server listens on 127.0.0.1:8765.
Validation calibration error 0.022, and 96-97% of answers are unchanged when options are shuffled.
Supports up to 255 options plus a reject probability for 'none of these fits'; the state is encoded once for cheap follow-up questions.
Scored 0.732 on JevBench public items versus Jev's 0.866; financial sentiment 0.927 and RAGTruth 0.813 in the project's benchmarks.

Watch out

Apache-2.0 code and weights. The README documents weak spots including near-chance date and number arithmetic, long multi-step rulebooks, ambiguity hedging, close-pair judging and English-only input.

Questions about this entry

Short answers drawn from the linked project or page.

Related terms

Glossary definitions related to this entry.

More Jev resources like this

WaterSheep

Apache-2.0 ModernBERT fine-tune that answers noul, choice, score and multi-label questions with a probability for every option. It serves Jev's POST /v1/systemone locally so TypeSafe's Python SDK runs unchanged, reporting 0.026 expected calibration error on its test split.

Jev Models & Reimplementations#community#calibration#modernbert

Community

SamratDuttaOfficial/WaterSheepPython

ruling

A local Jev-compatible server that reads Choice, Score and Noul answers from the logits of any MLX checkpoint or OpenAI-compatible endpoint with no training, serving POST /v1/systemone as a drop-in for the official TypeSafe SDK.

Jev Models & Reimplementations#community#mlx#calibration

Community

bradAGI/rulingPython

AutoJev

130

An open recipe for a 27B multimodal decision model trained with full-weight SFT on 73,000 examples over one H200, serving choice, noul and score on /v1/systemone with per-checkpoint provenance and released calibration plots.

Jev Models & Reimplementations#community#calibration#qwen

Community

denis-pplx/autojevPython

Vev

Jev-compatible decision models with vision input: LoRA fine-tunes of Qwen3.5 at 4B and 9B serving Choice, Score and Noul on the /v1/systemone wire format, with screenshots and photos placed beside text in the state. Runs on your own NVIDIA GPU; weights are non-commercial.

Jev Models & Reimplementations#community#vision#open-weights

Community

Xiaooolong/vevPython

Qwev

A training-free runtime that turns dense Qwen3 or Qwen3.5 checkpoints into a Jev-style Noul, Choice and Score service, sharing one state prefill across isolated questions and reading each answer from candidate-token logits.

Jev Models & Reimplementations#community#qwen#local-inference

Community

HopLee6/QwevPython

Ollaya

1.3k

A local daemon that pulls open decision models and serves them on TypeSafe's /v1/systemone, so a Jev client switches hosts with TYPESAFE_BASE_URL.

Jev Models & Reimplementations#community#rust#open-models

Community

ollaya-dev/ollayaRust

From the Jev guides

Original write-ups on what this entry is for.

Multimodal Decision Models: What Clef and d1 Change

Oct 12, 2026

Jev takes text only; Clef-omni adds image, audio, and video input and Liquid d1 adds images, each keeping the same typed-answer shape. Where that matters, where it does not, and the costs the vendors underplay.

Jev Models & Reimplementations#multimodal#open-models#decision-model

Read article

The Decision-Model Price War

Oct 11, 2026

One model in September, a dozen by October. Clef, Liquid d1, Perplexity, and OpenAI all ship the same typed-answer shape now. How to read the rival benchmarks and pick on your own task.

Jev Models & Reimplementations#decision-model#benchmarks#pricing

Read article

What Is Actually on Hugging Face

Oct 8, 2026

Hundreds of repositories carry the system-one tag, but the ecosystem is a dozen ideas in different sizes. A tour of the families, the licenses, and what a download does not include.

Jev Models & Reimplementations#open-models#hugging-face#local

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai