Skip to main content

Migrate from Jev

Super Decider accepts the same request format as TypeSafe's System One API, so most Jev integrations move by changing the endpoint, the model name and the key.

What to change​

SettingJevSuper Decider on ASI:Cloud
Endpointhttps://api.typesafe.ai/v1/systemone, or a gateway such as OpenRouterhttps://inference.asicloud.cudos.org/v1/systemone
modeljev-latest, jev-1.14.0, typesafe/jev-1.13super-decider. Any other name returns 400.
KeyTypeSafe or OpenRouter keyAn ASI:Cloud API key, as Authorization: Bearer ...

These stay the same: state, questions, the noul, choice and score types, the shapes of criteria, JSON in instructions and criteria, and the answer fields (noul, choice, probabilities, confidence, score, legend).

Differences to check​

AreaJevSuper Decider
ImagesAn assets map referenced from the request, behind an opt-in headerAn images array of up to 8 base64 data URLs next to state, no header. See Images.
Box questionsSupported in Jev 1.14Not supported. Returns 422.
Span questionsNot availablespan and spans copy text from the state with offsets. See Span.
ThinkingNo controlOff, automatic or on every question. See Thinking.
Limits131,072 tokens and up to 64 questions per request. A choice with more than 26 options is read in chunks and is slower. A score can have more than 10 levels.
UsageIncludes a cost fieldinput_tokens and output_tokens only. See the model catalogue for prices.
SDKsTypeSafe Python and JavaScript SDKsTypeSafe's SDKs target TypeSafe's endpoint. Call the HTTP API directly, or use an OpenAI client through chat completions.

Re-check after switching​

Different models are calibrated differently. Re-run your integration's test cases and re-tune your confidence thresholds on your own data before you rely on the old values. See setting thresholds.

The agent skill can make these changes for you. Ask your coding agent to switch the integration to Super Decider.