Native structured judgments
UsePOST /v1/systemone with a Codeflare Bearer key and a platform model with systemone support. This protocol is separate from chat. Requests contain only model, state, and questions.
cURL
Question types
Inputs can be strings, JSON objects, or arrays. Responses retain native
model, answers, and usage, including probabilities and confidence. The returned model identifies the actual serving version. The upstream service determines context and per-question limits.
Limits and failure handling
Streaming, background work, chatmessages, token counting, and generation-length parameters are unsupported. There is no chat-protocol conversion. Platform GET /v1/models still has an OpenAI envelope, so do not assume compatibility with the TypeSafe SDK’s model-list format.
Upstream 422, 429, and 529 statuses are retained with sanitized errors, without automatic retries. Missing or invalid usage enters unknown-cost reconciliation and must not be treated as a free call.
Check the console and official TypeSafe documentation for models, context limits, and prices. This guide does not pin upstream versions or prices.