Pacing batch jobs on the GPTpost agent API: limits and retries

Written by

in

Scripts and AI agents that schedule social posts in bulk run into rate limits sooner or later. Undocumented limits turn that into guesswork. GPTpost publishes its limits, so you can design jobs around them. Here is what they are and how to work with them.

The limits

  • About 120 requests per minute per API key.
  • About 30 writes per minute per API key (creating, scheduling or publishing posts, uploading media).
  • Every 429 response includes a Retry-After header.

The same limits apply to the REST agent API and to MCP tool calls.

Designing a batch job

Suppose you want to load a quarter of content: a few hundred posts across several accounts.

  1. Read once. Fetch the account list a single time at the start instead of before every post.
  2. Pace writes. One write every two to three seconds stays comfortably under 30 per minute.
  3. Honour Retry-After. On a 429, wait the number of seconds given, then continue. Do not hammer the endpoint.
  4. Back off on server errors. Retry 5xx responses with increasing waits.
  5. Use idempotency keys. Send a deterministic Idempotency-Key on every create, built from your own slot and account identifiers. If a request times out and you retry, no duplicate post is created.
  6. Keep local state. Record what was created, so a re-run skips finished work.

Errors that are not rate limits

A 422 means text is over a network’s limit, a 409 that approval is required, a 402 that a plan limit was reached. These will not succeed on retry, so log them and move on.

We use this ourselves

Our own 90-day social calendar was scheduled through this API with exactly these patterns. Full reference on the social media API page.

https://oraglegpt.org/use-cases/social-media-api/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *