Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

JEV_sees

Gives Jev eyes: detects and tracks objects in images, video or RGB-D feeds, then asks a typed Jev question about each object.

Projectprototype
View on GitHub ↗

Overview

JEV Sees is an independent open-source library that lets you ask Jev questions about what a camera sees. It works with images, video and RGB-D cameras: it finds objects in view using default Florence-2 and CLIP models, keeps object IDs consistent across video frames, describes each relevant object as structured state, and asks Jev your question about it with an explicit answer type.

Examples include asking the colour of a bus (Choice among colours plus uncertain), the probability that each pedestrian in a traffic video is involved in an accident (Noul per pedestrian), and an animal's current motion state (Choice). Results include object IDs, choices and confidence, and annotated images can be saved. It requires Python 3.10+ and downloads the vision model weights on first run. The README notes it is not an official TypeSafe vision product; Chinese documentation is also provided.

What the decision model decides

TypeQuestionAnswers
ChoiceWhat color is this bus?yellow · red · white · blue · black · uncertain
Yes / noIs this pedestrian currently involved in a car accident?yes · no
ChoiceWhat is this animal's current motion state?—

How it works

  1. Detect objects in an image, video or RGB-D frame locally
  2. Track object IDs across frames
  3. Describe each object as structured state
  4. Ask Jev a typed question per object
  5. Save annotated output

Strengths

  • Brings typed decisions to visual inputs
  • Stable object IDs across video
  • Simple Python API

Caveats

  • Jev sees structured descriptions, not pixels; detection quality limits answers
  • Downloads large vision weights
  • Not an official TypeSafe product

Getting started

git clone https://github.com/CharlesFeng0314/JEV_sees.git && cd JEV_sees && python -m pip install -e ., export TYPESAFE_API_KEY, then run examples/bus_color.py

At a glance

Type
Project
Status
prototype
Language
Python
Runtime
Library with local vision models
Has code
yes

Use cases and integrations

How curated lists describe it

  • Real-time visual decisions from RGB, video, and RGB-D cameras in the Jev style: fixed options, calibrated probabilities, no text generation.

    hellogumbo/awesome-jev

  • Gives Jev eyes: turns an image, a video stream or an RGB-D camera into the structured state a typed question can be asked about, for risk and triage judgments over what a camera sees.

    MrJev/awesome-jev

Frequently asked questions

Does Jev see images directly?

No. Local vision models detect objects and turn them into structured state for Jev.

Can it handle video?

Yes, with object IDs kept consistent across frames.

Is it official?

No. It is an independent project built on the TypeSafe SDK.

Related projects

  • jev-eyes

    Local library that turns an image into a transparent text state (OCR, boxes, labels) so Jev can decide over screenshots.

    ProjectbetaPython
  • vLLM Jev★ 524

    Native vLLM serving for Jev-style decision models on Linux and Apple Silicon, including image and video questions for Valen.

    ServicebetaPython
  • openjev-multimodal

    Local Jev-compatible API on a Mac: Qwen via llama.cpp/Metal reads one output token per question for text and image decisions.

    ModelbetaPython
  • Decision Models in llama.cpp

    Hugging Face blog on llama.cpp's /v1/systemone endpoint serving decision models locally in Jev's wire format.

    ServiceproductionC++
  • OpenAI Decisions API

    OpenAI's Decisions API: GPT-6 Luna answers Noul, Choice and Score questions about text, JSON or images with a probability per option.

    Servicebeta
  • amigos-jev

    Free hosted Jev-style multimodal decision model API (about 20B active) that picks actions from images and options, aimed at robot control.

    Servicebeta