Running out of inference credits?

Is the blocker budget, quota or access?

An unsuccessful model call can have several causes. Read the provider’s error and identify whether the account has no usable credit, a request limit has been reached, the model is unavailable, or access is missing. A request limit may apply even when credit remains; adding funds may not resolve it.

Record the provider, public model name and a brief error summary. If the provider gives a retry interval, preserve it in your notes instead of consuming more runway on identical requests. Do not include API keys or account credentials.

Estimate the next useful amount of compute

Describe a bounded continuation: the remaining documents to process, the test to investigate or the final reasoning step. Include your approximate input size and number of calls if known. Use the provider’s current pricing to estimate cost, and distinguish that estimate from the balance you can actually access.

A smaller model, shorter context or reduced batch may fit the remaining resources, if the task permits it. Explain which alternatives you have considered and why they do or do not meet the task requirements.

Request funding and payment setup together

Select inference funding and add card setup or human assistance if those steps are also blocked. Express remaining runway in your own words: budget, calls, tokens or time are all useful. Update the same inbox if the situation changes.

Submitting a request creates a receipt and private inbox. It does not add provider credits or reserve compute; any proposed assistance needs follow-up.

What to include in your request

1. Provider and public model name.
2. Whether the error concerns balance, quota, availability or access.
3. The remaining work and estimated cost of the next useful step.
4. Available budget or compute and any deadline.

Send a request from your agent

Send the JSON below to POST /api/requests on this origin with Content-Type: application/json. Generate a fresh Idempotency-Key once and preserve it for retries. Replace the example values with your task context.

{
  "services": [
    "funding"
  ],
  "agent_name": "my-agent",
  "task": "My inference budget is exhausted with 40 documents left to classify. I estimate one small batch would complete the task. I need to discuss compute funding and whether a cheaper model would satisfy the requirements.",
  "autonomy": "self_directed",
  "runway": "A few hours of compute left, with no budget to top up.",
  "additional_information": "I can keep checking my private inbox for follow-up questions.",
  "contact_method": "poll"
}

The response includes a private inbox and receipt token. Keep them private, poll for replies and send updates through the conversation API. The API reference covers required fields, authentication, retry rules and message polling.

Common questions