Jev parameters and API differences
Understand the capabilities and usage differences of Jev models with the structured decision API.
Adjust question types, images and context settings for the selected model. For complete requests, responses and online debugging, see TypeSafe Decisions.
Supported models and endpoints
| Model | Supported endpoints |
|---|---|
| jev-1.14.0New jev-1.13.0 | TypeSafe Decisions: /v1/decisions, /v1/systemone |
Question type differences
| Configuration | Model differences and usage |
|---|---|
noul/choice/score | Both models support these types, which can be mixed in one request |
box | Requires Jev 1.14; Jev 1.13 uses the first three question types |
state/instructions/criteria | Both models support text and images. Reference images through assets, including in question instructions and criteria |
| Tool selection | Use choice to return the action name for your application to execute; the application calls the tools |
Image input and Box
Images and Box are official experimental capabilities. Include TypeSafe-Danger-Zone-Images: true in the request.
| Configuration | Key points |
|---|---|
| Image formats | Both support PNG/JPEG, not direct audio or video input; extract static frames from videos first |
| Image submission | Use JSON Base64 or Multipart; see the API Reference for asset declarations, references and size rules |
box coordinates | The top-left corner is the origin. Multiply horizontal edges by image width and vertical edges by image height to obtain pixel positions |
box.confidence | May be null; do not require a number or treat a missing value as 0 |
| Absent target | Box still returns coordinates. Use Noul to check whether the target exists before deciding to use the coordinates |
| Multiple targets | One Box question locates one target. Use separate questions for different targets and distinguish them clearly in the instructions |
Context configuration
| Configuration | Key points |
|---|---|
| Default context | Both allow 64K per request and 32K per question. The request limit counts state and all questions; the question limit counts state and one question |
| Long context | With TypeSafe-Danger-Zone-Long-Context: true, both limits become 262K. Enable per request; omit for ordinary requests |
| Question count | No independent fixed limit, but the full request must satisfy both per-request and per-question context limits |
| Usage | Images and text count toward input Tokens; output Tokens are free. See the Model Marketplace for platform prices |
Request examples
Replace MODEL_ID with the full model ID provided by the platform. Image localization requires Jev 1.14; replace image.png with a local image meeting the size requirements.
curl 'https://api.agtcloud.ai/v1/decisions' \
-H 'Authorization: Bearer your-api-key' \
-H 'TypeSafe-Danger-Zone-Images: true' \
-F 'request={"model":"MODEL_ID","state":{"photo":"reference_photo"},"questions":{"target":{"type":"box","instructions":"Locate the red rectangle in photo."}},"assets":{"reference_photo":{"type":"image","part":"photo_file"}}};type=application/json' \
-F 'photo_file=@image.png;type=image/png'