Calls OpenAI via the Responses API and merges structured output into the payload.
Calls OpenAI via the Responses API and merges structured output into the payload.
When the flow reaches this step, JOrchestrator calls the configured model using credentials from Node Configuration, then exposes the response fields listed below to downstream steps.
OpenAI is a AI step in the AI group. AI steps call language or media models. Use them to generate text, classify content, or produce assets that later steps publish or send.
Typical reasons teams add OpenAI to a flow:
Common patterns on the canvas:
General — model, temperature, max tokens. Design — available fields, tools, output format, response JSON shape, and prompt.
Enable file_search or MCP tools on Design when the model needs external context.
The designer shows these tabs for this node:
| Tab | Purpose |
|---|---|
| Info | Plain-language description and output payload fields for downstream steps |
| General | Canvas label and read-only node ID |
| AI | Optional assistant to help draft settings from plain language |
| Design | Node-specific configuration fields |
These fields are managed by the designer or runtime — you do not type into them directly:
| Property key | Label | Type | Required | Visible when | Notes |
|---|---|---|---|---|---|
_nodeId | Node ID | node-id | No | — | — |
_payloadFields | Available Fields | payload-panel | No | — | — |
| rawJsonShape | OpenAI response JSON | raw-json-shape-editor | No | outputFormat === 'raw_json' | { "title": "", "introtext": "", "customKey": "" } |
| Property key | Label | Type | Required | Visible when | Notes |
|---|---|---|---|---|---|
label | Label | text | No | — | Node label |
model | Model | text | No | — | Model name (e.g., gpt-4, claude-3, etc.) |
temperature | Temperature | number | No | — | 0.7 |
maxTokens | Max Tokens | number | No | — | 1000 |
| Property key | Label | Type | Required | Visible when | Notes |
|---|
| agentPreset | Agent preset | select (transform \| research \| classifier \| editor \| custom; default: transform) | No | — | Transform (default) — Single-shot transform — no agent role or tools. Research agent — Web search (OpenAI) + structured JSON with citations. Classifier — Categorize or route content into structured labels. Editor — Editorial assistant — article-shaped output. Custom — Write your own system prompt below. |
systemPrompt | System prompt | textarea | No | — | Agent role and constraints (optional). Applied separately from the task prompt below. Defines how the model behaves (research agent, editor, classifier, etc.). The task prompt below describes what to do with upstream data this run. |
|---|---|---|---|---|---|
openaiTools | Tools to include | openai-tools | No | — | — |
openaiVectorStoreIds | Vector store IDs (file search) | openai-vector-store-ids | No | openaiToolsIncludes:file_search | vs_… one per line or comma-separated |
openaiMcpServerLabel | MCP server label | text | No | openaiToolsIncludes:mcp | Shown in tool calls (required for MCP) |
openaiMcpServerUrl | MCP server URL | text | No | openaiToolsIncludes:mcp | https://… (required unless connector ID is set) |
openaiMcpConnectorId | MCP connector ID | text | No | openaiToolsIncludes:mcp | Optional ChatGPT / connector id if not using URL |
outputFormat | Output format | select (legacy | image | article | raw_json | email; default: article) | No | — | — |
prompt | Prompt | textarea | Yes | — | Enter your OpenAI prompt here… Drag fields from Available Fields into the prompt. |
Credentials, API keys, mailers, SMTP, and integration defaults are configured in the CMS plugin under Node Configuration, not per step on this node. Link the flow to a domain before testing CMS-backed fields.
After this step runs, later nodes can reference these fields using placeholders such as {{openai.parsed}}.
| Field key | Type | Description |
|---|---|---|
parsed | object | Parsed JSON object extracted from the LLM response. |
title | string | Title text for the created or updated record. |
introtext | string | Article introduction / teaser HTML. |
fulltext | string | Article main body HTML. |
response | string | Raw text response from the LLM or API. |
content | string | Main text content from the LLM response. |
messageId | string | Provider message or email identifier. |
usage | object | Token usage metadata from the LLM API. |
prompt) must be filled before the flow can be saved.Other steps in the same area of the palette:
ai-anthropic) — AIai-cohere) — AIai-google-gemini) — AIai-groq) — AIai-meta-llama) — AI