Skip to main content

​
OpenAI

Link an OpenAI account where the node shows Third-party account.

​
Chat

Single-turn or stateless prompt to a chat model. Supports file attachments (images or PDFs), optional web search, and JSON Schema structured output. Inputs
  • Prompt — User message text.
  • Model — Includes GPT-5.x / GPT-4.x “standard” models and reasoning models (o3, o4-mini, etc.).
  • Reasoning models — Show reasoning effort (low / medium / high) instead of temperature/web search.
  • Non-reasoning models — Web search toggle; temperature slider (0–2, default 1).
  • Structured output — When enabled, provide a JSON Schema string whose top-level type must be "object" with a properties map. Output becomes a structuredResponse object you can destructure on the graph. The editor can suggest adding schema properties as outputs.
  • System prompt — Optional (config).
  • Files — Optional list of files (images or PDFs) sent with the prompt.
Outputs
  • Default: response (string).
  • Structured output mode: structuredResponse (object matching your schema).

​
Classify Text

Inputs
  • Model — Same picker family as other OpenAI nodes (default gpt-5-mini on this node).
  • text — String to classify.
  • Include “other” category — Adds an escape hatch if nothing matches.
  • Allow multiple categories — Single selectedCategory vs list selectedCategories.
  • categories — Repeatable key (what the model outputs) + optional description (guidance for the model).
Outputs
  • selectedCategory or selectedCategories — Chosen label(s).
  • One boolean per category — driven outputs (is_*) such as Is “invoice” for each key you define.
  • If “other” is enabled: isOther.

​
Extract Text

Inputs
  • Model; text source.
  • features — Repeatable strings naming each field to pull out (e.g. "invoice total", "due date").
Outputs
  • Driven outputs — One string per feature (“Extracted Feature 1…”) you can rename by editing the feature list.

​
Produce Text

Runs the model once per prompt in parallel conceptually; each output is independent. Inputs
  • Model; prompts — Repeatable string prompts.
Outputs
  • Driven outputs — Produced Text 1…N, one string per prompt entry.

​
Yes/No Question

Inputs
  • Model; question — Yes/no style question; additionalText — Optional context (e.g. the value being judged).
Output
  • decision — Boolean (true = yes / affirmative).

​
Generate Image

Inputs
  • prompt; model — DALL·E 2, DALL·E 3, or GPT Image 1.
  • size — Depends on model (e.g. 256–1024 squares, DALL·E 3 aspect ratios, GPT Image sizes including auto).
  • quality / style — When the model supports them (DALL·E 3 quality & style; GPT Image quality).
  • GPT Image 1 only: background (auto / transparent / opaque), output format (png/jpeg/webp), output compression (0–100 for lossy formats).
Output
  • image — Generated file.

​
Text to Speech

Inputs
  • text; model — tts-1, tts-1-hd, or gpt-4o-mini-tts.
  • voice — Enum (alloy, nova, etc.).
  • instructions — Shown only for gpt-4o-mini-tts; steers delivery style.
  • speed — 0.25×–4×.
  • response_format — mp3, opus, aac, flac, pcm.
Output
  • audio — File.

​
Transcribe Audio

Inputs
  • audio file; model — whisper-1, gpt-4o-transcribe, or gpt-4o-mini-transcribe.
Output
  • text — Transcript string.