n8n-nodes-openai-structured-outputs

This is a community node for n8n that enables you to extract structured JSON from unstructured text by leveraging OpenAI’s Structured Outputs feature with a user‑provided JSON Schema.

---

Installation

You can install n8n-nodes-openai-structured-outputs in two ways—via npm or directly from the n8n app.

---

1. Install via npm

1. In your project (or globally), run:

   npm install n8n-nodes-openai-structured-outputs
   

2. Restart your n8n instance to load the new node.

---

2. Install from the n8n App (Community Nodes)

Note: Only the n8n instance owner can install and manage community nodes.

1. In n8n, go to Settings → Community Nodes.
2. Click Install.
3. Click Browse—n8n will show npm packages tagged n8n-community-node-package.
4. Search for and select n8n-nodes-openai-structured-outputs.
   • To install a specific version, append @<version> (e.g. n8n-nodes-openai-structured-outputs@*.*.*).
5. Check “I understand the risks of installing unverified code from a public source”, then click Install.
6. After installation, the OpenAI Structured Output node appears under Community Nodes → Utility.

---

3. Uninstall, Upgrade or Downgrade

• Uninstall: In Settings → Community Nodes, click the three-dot menu next to OpenAI Structured Output and choose Uninstall.
• Upgrade to Latest: If a new version is available, an Update button appears next to the node—click it to upgrade.
• Install Specific Version: Uninstall first, then reinstall using the exact npm package name with version (e.g. n8n-nodes-openai-structured-outputs@*.*.*).
• Downgrade: Same as installing a specific older version.

---

For full details and additional community-node management guidance, see the official docs:
https://docs.n8n.io/integrations/community-nodes/installation/

---

Operations

Extract JSON

Use the Extract JSON operation to send text to the OpenAI Chat Completions endpoint with your JSON Schema injected into response_format. The node will:

1. Call POST /v1/chat/completions with:

   • model (selectable)
   • messages: system prompt + user text
   • response_format: { type: "json_schema", json_schema: { name, schema, strict: true } }
   • optional parameters like temperature

2. Receive the AI’s response, automatically:

   • Validate HTTP errors (4xx/5xx)
   • Parse the first choice’s message.content as JSON
   • Validate the parsed JSON against your schema (using AJV with coercion, defaults, and stripping extra fields)
   • Output:

     {
         "parsed": {
             /* your valid JSON */
         },
         "errors": [
             /* AJV errors, if any */
         ],
     }
     

Node Parameters

Parameter	Name	Type	Default	Description
Model	model	options	gpt-4o	OpenAI model to use (e.g. gpt-4o-mini, o3-mini, etc.)
JSON Schema	jsonSchema	json	""	A valid JSON Schema (injected into response_format)
Text	text	string	""	Input text from which to extract structured data
Temperature	temperature	number	1	Sampling temperature (0–1)

---

Credentials

This node uses the built‑in OpenAI API credential:

1. API Key (required)
2. Organization ID (optional)
3. Base URL (defaults to https://api.openai.com/v1)

Configure these under Settings → API Credentials → OpenAI in n8n.

---

Compatibility

• n8n version: ≥ 1.x with Community Nodes API v1
• Node.js: ≥ 18.10
• AJV version: 8.x (bundled)

---

Usage

1. Add the OpenAI Structured Output node to your workflow.
2. Select Extract JSON.
3. Fill in your Model, paste or compose your JSON Schema, and supply the Text.
4. Connect an output to inspect:
   • parsed: the coerced, default‑filled, extra‑fields‑removed object
   • errors: any AJV validation errors

You can chain this into further nodes (e.g., write to a database, send an email) knowing the JSON conforms to your schema.

---

JSON Schema Reference

We leverage the industry‑standard JSON Schema draft (see official docs):
https://json-schema.org/understanding-json-schema/

Notes on Keyword Support

OpenAI’s Structured Outputs supports many common JSON Schema keywords, but also has explicit limitations. Below is a summary of what works (✔) and what does not (✘) for each data type:

String

• ✔ type
• ✔ enum
• ✔ const
• ✔ contentEncoding (if supported)
• ✔ contentMediaType (if supported)
• ✘ minLength, maxLength, pattern, format

Number & Integer

• ✔ type
• ✔ enum
• ✔ const
• ✔ exclusiveMinimum, exclusiveMaximum
• ✘ minimum, maximum, multipleOf

Boolean

• ✔ type
• ✔ enum
• ✔ const

Object

• ✔ type
• ✔ properties
• ✔ required (mandatory): an array that must list all property names declared under properties.
• ✔ additionalProperties (mandatory): must be explicitly set (false or a subschema) to control extra fields.
• ✔ dependencies (if supported)
• ✔ patternDependentSchemas (if supported)
• ✘ patternProperties, unevaluatedProperties, propertyNames, minProperties, maxProperties

Array

• ✔ type
• ✔ items
• ✔ additionalItems
• ✔ prefixItems (if supported)
• ✘ unevaluatedItems, contains, minContains, maxContains, minItems, maxItems, uniqueItems

General (Multiple Types)

• ✔ enum
• ✔ const
• ✔ anyOf
• ✔ allOf (not explicitly unsupported)
• ✘ oneOf, not (unsupported or untested)
• ✔ if, then, else (if supported)
• ✔ description, title
• ✘ default
• ✘ $ref, $defs/definitions (no external references)

Tip: Schema reuse via $defs inside the same document may work, but external $ref is not supported.

---

Resources

• n8n Community Nodes: https://docs.n8n.io/integrations/community-nodes/
• OpenAI Chat Completions: https://platform.openai.com/docs/guides/chat/introduction
• JSON Schema Official: https://json-schema.org/understanding-json-schema/

---

Version History

• 0.1.0 (2025‑04‑22)
  • Initial release: Extract JSON via OpenAI Structured Outputs with full AJV validation, coercion, defaults, and extra‑field removal.

---

Created by Crucerlabs · MIT License