Skip to content

Batch inference

Trussium exposes provider-neutral lifecycle metadata for asynchronous batch jobs:

  • POST /v1/batches
  • GET /v1/batches/{batch_id}
  • POST /v1/batches/{batch_id}/cancel

The initial OpenAI integration accepts only /v1/chat/completions batches with a caller-owned provider input_file_id. Input and output files remain at the provider: Trussium does not upload, download, persist, or log their contents.

Batch requests are for offline work and use the provider's fixed 24-hour completion window. Arbitrary endpoint proxying, file uploads, result downloads, webhooks, retries, and local scheduling are outside this capability.