OpenDecision
Local decision engine on zero-shot ModernBERT and SigLIP2 answering Choice, Noul, Score and Relation over text, documents and images.
Overview
OpenDecision is an open-source, locally run decision engine that describes itself as an equivalent of Jev. You send text, JSON, a document or an image with typed questions and get structured answers: Choice selects one option, Noul scores a statement or compares true/false descriptions, Score places state on an ordered scale, and Relation reports supports, contradicts, unknown or conflicted. Document processing retrieves evidence and source passages.
The default backend is MoritzLaurer's ModernBERT-large zero-shot model, with DeBERTa-v3-large optional, and image decisions use SigLIP2. It offers a Python API, a FastAPI server and a TypeSafe SDK-compatible endpoint, under Apache 2.0. Image examples such as wildlife alerts, overflowing bins and damaged fences map answers to simulated actions, with median call times of 155-318 ms on an Apple M3 using a yes/no threshold of 0.5 and a minimum winning Choice probability of 0.6.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is an animal visible? | yes · no |
| Choice | What condition is the bin in? | — |
| Choice | Relation between claim and evidence | supports · contradicts · unknown · conflicted |
How it works
- Send text, JSON, a document or an image with typed questions.
- Run zero-shot NLI or SigLIP2 locally.
- Apply thresholds (e.g. 0.5 yes/no, 0.6 choice).
- Map answers to actions such as sound_alarm or request_collection.
Strengths
- Fully local with small zero-shot models.
- Images and documents with evidence passages.
- TypeSafe SDK-compatible endpoint.
Caveats
- Zero-shot NLI models are less capable than LLM-based decision models.
- Image actions in examples are simulated.
At a glance
- Type
- Service
- Status
- beta
- Language
- Python
- Runtime
- library, FastAPI server
- Has code
- yes
- License
- Apache-2.0
- Stars
- 58
Use cases and integrations
How curated lists describe it
Open alternative: a local semantic decision engine that describes itself as the open-source equivalent of Jev, answering `Choice`, `Noul`, and `Score` questions from structured state and documents without a hosted call.
yibie/awesome-jevOpenDecision is an open-source semantic decision engine like typesafe's jev.
wh000wh000/awesome-jev-liveOpenDecision is an open-source semantic decision engine like typesafe's jev.
hellogumbo/awesome-jevOpen-source semantic decision engine: state, a question in natural language, and answer criteria in; a structured decision out.
MrJev/awesome-jev
Frequently asked questions
Which models does it use?
ModernBERT-large zero-shot by default, DeBERTa-v3-large optionally, and SigLIP2 for images.
Can it read documents?
Yes, with evidence retrieval and source passages.
Does it work with the TypeSafe SDK?
Yes, it provides a compatible endpoint.
Related projects
- Dohnuts★ 36
Small multimodal System One models that choose, judge or score text, documents and images in one forward pass.
ModelresearchPython Local library that turns an image into a transparent text state (OCR, boxes, labels) so Jev can decide over screenshots.
ProjectbetaPythonPermission-aware document library using Jev or Cloudflare Clef for hierarchical search, passage scoring and automatic filing.
LibrarybetaTypeScriptFree hosted Jev-style multimodal decision model API (about 20B active) that picks actions from images and options, aimed at robot control.
ServicebetaComfyUI nodes for Jev-style noul, choice and score decisions over images and text using a local imajev adapter on Qwen3.5.
ProjectbetaPythonResearch code for an image-and-text System One model that reads option-letter logits from a multimodal LLM, with training and calibration.
ProjectresearchPython