Gutsy
Apache-2.0 local decision model (0.8B Qwen fine-tune on llama.cpp) behind the Jev request shape, calibrated and CPU-only.
Overview
Gutsy provides local decision models with calibrated probabilities. You send a state (text or JSON) and typed questions (yes/no, choice, score) and get a probability for every option back, with no text generation, network calls or per-call cost. The API follows the Jev-style decisions format, so existing clients mostly need only a new base URL.
The model, gutsy-0.8b, is a fine-tune of Qwen3.5-0.8B available as Q8_0 (775 MB) or a faster Q4_K_M (505 MB), served by gutsy-inference on llama.cpp on an ordinary CPU. The README reports validation calibration error of 0.022, determinism on the same machine and build, 96 to 97% of answers unchanged when options are shuffled, state encoded once so extra questions cost only their own tokens, up to 255 options, and a reject probability for "none of these fits". It is Apache-2.0.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What should support do next? | — |
| Yes / no | Yes/no questions about the state | yes · no |
| Score | Ordered rating questions | — |
How it works
- Install gutsy-inference and download the model
- Start the local server
- POST state and questions to /v1/systemone
- Use probabilities, including the reject option
Strengths
- CPU-only, private and free per call
- Reported good calibration
- Reject option for no-fit cases
- Jev-compatible request shape
Caveats
- Small model limits accuracy on hard tasks
- Requires llama-cpp-python
Getting started
pip install -e gutsy/gutsy-inference, hf download kouhxp/gutsy --include "gutsy-0.8b-v04*", then gutsy-inference serve (http://127.0.0.1:8765).At a glance
- Type
- Service
- Status
- beta
- Language
- Python
- Runtime
- llama.cpp local server
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Local decisions: an Apache-2.0 model that answers typed questions with calibrated probabilities on CPU, with no GPU and no hosted API.
yibie/awesome-jevLocal decision model behind the Jev request shape: no generation, no network calls, no per-call cost.
MrJev/awesome-jev
Frequently asked questions
Does it need a GPU?
No, it runs on an ordinary CPU.
Is it Jev-compatible?
It follows the Jev-style request format; clients mostly need a new base URL.
How well calibrated is it?
The README reports validation calibration error of 0.022.
Related projects
- cbjev★ 4
Self-hosted encoder model giving Jev-compatible choice, score and noul answers in one pass, pitched as a better-calibrated Laya successor.
ProjectresearchPython 11 MB CPU-only model that answers choice, noul and score questions with fitted calibration and no LLM at inference.
ProjectresearchPython- kev★ 9,079
Family of small Jev-like decision models you can train and run yourself, with a server compatible with TypeSafe's SDK.
ServicebetaPython Open 596M and 4B decision models that answer Choice, Score and Noul in one pass locally, with calibrated confidence for escalation.
ServicebetaPythonApache-2.0 118M multilingual decision model for Choice, Score and Noul on the System One API, with conformal abstain.
ModelbetaPythonOpen Apache-2.0 System One decision models (4B, 9B, 27B on Qwen) with trainer, data builders, evals and every run record.
ServiceresearchPython