Skip to content

OpenAI Batches API: using the VIDAI Control Plane as the backend

The Batches API is store-and-forward: you upload a JSONL of requests, OpenAI processes them over up to 24 hours, and you download the results. The control plane's shipping surface is per-request (each call is an interactive proxy hop) — the Batches state machinery (file upload, batch create, batch retrieve) is not proxied today.

If your traffic is genuinely batch-shaped, you have two paths:

  • Submit batches directly to OpenAI and route your interactive traffic through the control plane. Two auth paths, two cost pools — but batches get their own steep OpenAI discount (~50% off list) that you'd sacrifice by interactive-routing them.
  • Convert batch-shaped work to a rate-limited interactive loop through the control plane. Rate limits and cost tracking apply the same way; you lose the batch discount but gain full control-plane observability + guardrails + chargeback attribution.

Which is right depends on volume + cost sensitivity. For compliance-heavy work (batches whose prompts need guardrail enforcement), the interactive-loop path is the safer default — control-plane guardrails don't apply to batches that never touch us.

What we're considering

Batch-API proxying is on our roadmap. Priority again depends on demand — raise an issue if your project needs it.

If something's off

Raise an issue at github.com/vidaiUK/vidai-quickstart/issues describing the batch pattern you'd like to route. We'll respond with a timeline or a workaround.

Where to go next