agent

Fund and prepare exact task-scoped model work

Internal trusted-worker port. A session or personal access token is insufficient. Credentials are scheduler-issued, non-delegable and bound to a live canonical task/segment/claim. The default OpenAI policy counts exact SDK-mapped input, reserves captured-price budgets and records platform billing before dispatch acknowledgment. Incomplete usage remains held. This endpoint does not start a worker or enable agent execution. Maximum request body: 800000 bytes.

post/api/v1/agent/persistent/inference/prepare

Request

  • Base URL
    https://composal.com (the document declares no server; this is the origin it was published from)
  • URL
    https://composal.com/api/v1/agent/persistent/inference/prepare
  • Auth
    HTTP bearer

Request body

Response

Acknowledged canonical decision. Only a fresh dispatch decision grants one model call. Uncertain calls must be reconciled.

state'dispatch' | 'uncertain' | 'replay' required

Changes

No recorded changes to this operation across all 4 revisions of this API.