amigos-jev
Free hosted Jev-style multimodal decision model API (about 20B active) that picks actions from images and options, aimed at robot control.
Overview
AmigosJev is a Jev-style, natively multimodal decision model built by amigos, with about 20B active parameters, offered as a free API with no sign-up. It uses the model's next-token probability distribution directly as the decision output: it generates no text, answers always fall within the options you provide, and probabilities are renormalized over that option set.
It accepts text or JSON state plus up to four images and returns choice, score or yes/no answers with probabilities in 100 to 500 ms per the README. The main pitch is robot control: the robot's camera view plus a state description go in, the options are action commands (forward, turn left, grab, stop), and low confidence can trigger a stop or human decision. Demos include a live Snake game and a sketch-guessing game. Documentation is in Chinese.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which action should the robot take next given this camera view? | forward · turn left · grab · stop |
| Choice | Which of four options does this drawing show? | — |
How it works
- Send state text plus up to 4 images and options to /v1/systemone
- Model does one forward pass per question
- Probabilities are renormalized over your options
- Execute the chosen action, or stop when confidence is low
Strengths
- Free with no application
- Native image input
- Outputs restricted to your action space
Caveats
- Hosted by a third party with no stated SLA or licence
- Not TypeSafe's Jev; quality claims are the provider's
- Docs in Chinese
Getting started
Call https://jev.amigos-robot.com/v1/systemone (docs and demos at https://jev.amigos-robot.com/).At a glance
- Type
- Service
- Status
- beta
- Runtime
- hosted API
Use cases and integrations
How curated lists describe it
A free multi-modal jev API for everyone.
hellogumbo/awesome-jev
Frequently asked questions
Is this TypeSafe's Jev?
No, it is an independent Jev-style multimodal model with a compatible-looking endpoint.
Can it see images?
Yes, up to four images per request.
Does it cost anything?
The README says it is completely free with no application needed.
Related projects
Independent open model on Molmo2-ER that answers spatial questions by choosing from fixed options with one shared classifier head.
ProjectresearchPythonOmniJev: omni-modal 4B decision model answering typed questions about images, video, screens and robot scenes in one pass.
ModelresearchPythonComfyUI 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.
ProjectresearchPythonRepository serving CLM-8B, a contrastive System One model behind a TypeSafe-compatible API, aimed at humanoid robot decisions.
ProjectresearchPythonOpen project training a vision-enabled Jev-like model (0.8B, 9B) that judges dynamic candidate sets over images with Choice, Noul and Score.
ProjectresearchPython