OpenAI

The openai provider calls the OpenAI API. Its generator adapter also serves OpenAI-compatible endpoints; set url to point it elsewhere.

What It Provides#

The generator sends images as image_url parts. Other media, such as audio or PDFs, is rejected before the request is sent.

Credentials#

api_key, defaulting to OPENAI_API_KEY. Store it as a secret reference such as ${secret:openai.api_key}.

Example#

{
  "generators": {
    "answers": {
      "provider": "openai",
      "model": "gpt-5.6-luna",
      "api_key": "${secret:openai.api_key}",
      "max_completion_tokens": 1024,
      "reasoning_effort": "none"
    }
  }
}

Generation Options#

For models that require a completion budget, set max_completion_tokens instead of max_tokens. The budget covers visible output and reasoning tokens, and provider token quotas reserve it. The two limits are mutually exclusive; max_tokens defaults to 256 when neither is set. Antfly does not infer which limit a model accepts from its name.

reasoning_effort (none, minimal, low, medium, high, xhigh, or max, depending on the model) and sampling options such as temperature are opt-in and forwarded unchanged. Omit any the selected model does not support. These options apply to the openai provider only. See the OpenAI Chat Completions reference for parameter support.

Function tools work with any model that supports function calling; the adapter does not make a model tool-capable.

Extraction#

Use openai as an extraction provider: name the model and the schema, and Antfly runs a generated extractor on that model. As an extractor producer on an enrichment:

{
  "type": "extractor",
  "config": {
    "provider": "openai",
    "model": "gpt-4.1-mini",
    "api_key": "${secret:openai.api_key}",
    "schema": {
      "entities": ["person", "company"],
      "relations": [{"type": "works_for"}]
    }
  }
}

The schema takes entities and relations. Admitted model families: gpt-4o, gpt-4.1, and gpt-5 (including point releases such as gpt-5.1), each with optional mini or nano tiers and dated snapshots, plus o3 and o4-mini. Other models, and task variants such as audio, image, search, or embedding models, are rejected when the config is saved. Models in the gpt-4o and gpt-4.1 families score extractions with token log-probabilities; other models use verdict-only scores.

To set instructions, output mode, probabilities, or batching, use the generated form instead, which takes the same generator config under generator:

{
  "method": "generated",
  "generator": {"provider": "openai", "model": "gpt-4.1-mini", "api_key": "${secret:openai.api_key}"},
  "probabilities": "verdict_only",
  "instructions": "Only extract people named in the signature block.",
  "schema": {"entities": ["person"]}
}

The generated form, with inputs added, is also what /ai/v1/extract accepts.

Related#

• Provider rate limits
• Inference Components and Providers