LLM

Large Language Model API via AIG Gateway

Actions

Steps this node can perform. Add one from Add a step, then pick the action.

Text Chat

Generate text responses using language models

Settings

Field	Type	Required	Description
Model	Choice (loaded from your account)	Yes	Model identifier in provider/model format (e.g., 'openai/gpt-5.1'). Dynamically populated from AI Gateway. Default: openai/gpt-5.2
Messages	List	Yes	Array of conversation messages
Response Format	Choice		Controls the output format. 'text' for free-form, 'json_object' for valid JSON, 'json_schema' for schema-constrained JSON. Options: Text, JSON Object, JSON Schema Default: text
Schema Definition	Text	Yes	A JSON Schema definition that constrains the LLM output structure. The root must be {"type": "object"} - paste a schema, not an example of the output. Example: {"type":"object","properties":{"title":{"type":"string"}},"required":["title"]}
Temperature	Number		Controls randomness (0.0-2.0) Default: 0.7
Max Tokens	Number		Maximum tokens to generate Default: 1000
Frequency Penalty	Number		Penalize frequent tokens Default: 0
Presence Penalty	Number		Penalize presence of new tokens Default: 0
Stop Sequence	Text		Stop generation when sequence is encountered
Reasoning Effort	Choice		Hint for supported reasoning models Options: Low, Medium, High
Verbosity	Choice		Desired verbosity of text response Options: Concise, Balanced, Detailed
Retry Configuration	Object		Configure retry behavior for failed requests
Custom Tags	List		Optional custom tags for cost tracking (e.g., 'team:api', 'feat:summarize', 'env:prod')

Output

Later steps use this data by picking this step in the $ picker. For example, model_used is $.<step ID>.output.model_used.

{
  "model_used": "string",
  "response": {
    "choices": [
      {
        "finish_reason": "string",
        "index": 0,
        "message": {}
      }
    ],
    "created": 0,
    "id": "string",
    "model": "string",
    "object": "chat.completion",
    "usage": {
      "completion_tokens": 0,
      "prompt_tokens": 0,
      "total_tokens": 0
    }
  },
  "success": false,
  "tokens_used": 0
}

Example values show the shape of the data, not real values. See Using data from earlier steps.

Field descriptions

Field	Type	Description
model_used	Text	The model that was actually used for the request
response	Object	Raw unified LLM response payload
response.choices[]	List	A list of chat completion choices
response.choices[].finish_reason	Text	The reason the model stopped generating tokens
response.choices[].index	Number	The index of the choice in the list
response.choices[].message	Object	The message generated by the model
response.choices[].message.content	Text	The contents of the message (raw string, including JSON string when using structured output)
response.choices[].message.parsed_content	Any	Parsed JSON object when response_format is json_object or json_schema. Allows downstream nodes to access fields directly (e.g., parsed_content.name) without manual JSON parsing. Omitted when response_format is text or when JSON parsing fails.
response.choices[].message.role	Text	The role of the message author (always 'assistant' for responses)
response.choices[].message.schema_validation_errors[]	List	List of schema validation errors when the LLM output does not conform to the provided JSON Schema. Only present in json_schema mode when validation fails. The parsed_content field is still populated even when validation errors exist.
response.created	Number	The Unix timestamp (in seconds) of when the response was created
response.id	Text	A unique identifier for the response
response.model	Text	The model used for the response
response.object	Text	The object type, which is always 'chat.completion'
response.usage	Object	Usage statistics for the completion request
response.usage.completion_tokens	Number	Number of tokens in the generated completion
response.usage.prompt_tokens	Number	Number of tokens in the prompt
response.usage.total_tokens	Number	Total number of tokens used in the request
success	Yes / No	Whether the request was successful
tokens_used	Number	Total tokens consumed by the request

Generate Image

Generate images from text descriptions

Settings

Field	Type	Required	Description
Model	Choice (loaded from your account)	Yes	Provider-prefixed model id from AI Gateway Default: openai/gpt-image-1.5
Prompt	Text	Yes	A text description of the desired image(s)
Number of Images	Number		The number of images to generate Default: 1
Size	Choice		The size of the generated images Options: Auto, 1024x1024, 1024x1536, 1536x1024 Default: auto
Quality	Choice		The quality of the image Options: Low, Medium, High Default: medium
Output Format	Choice		The format in which the generated images are returned Options: Base64 JSON, File Default: file
File Name	File	Yes	The configuration for the output file. The file extension will be determined by the output_type field
Output Type	Choice		The image file format for the output Options: PNG, JPEG Default: png
Retry Configuration	Object		Configure retry behavior for failed requests
Custom Tags	List		Optional custom tags for cost tracking (e.g., 'team:api', 'feat:summarize', 'env:prod')

Output

Later steps use this data by picking this step in the $ picker. For example, created is $.<step ID>.output.created.

{
  "created": 0,
  "data": [
    "string"
  ]
}

Example values show the shape of the data, not real values. See Using data from earlier steps.

Field descriptions

Field	Type	Description
created	Number	Unix timestamp (seconds) when images were generated
data[]	List	List of generated images

Edit Image

Edit or transform existing images using AI (1-10 reference images, max 4MB each)

Settings

Field	Type	Required	Description
Model	Choice (loaded from your account)	Yes	Provider-prefixed model id from AI Gateway Default: openai/gpt-image-1.5
Reference Images	List	Yes	Reference images to edit (1-10 images, max 4MB each)
Prompt	Text	Yes	Text description of the desired edit or transformation
Number of Images	Number		The number of images to generate Default: 1
Size	Choice		The size of the generated images Options: Auto, 1024x1024, 1024x1536, 1536x1024 Default: auto
Quality	Choice		The quality of the image Options: Low, Medium, High Default: medium
Output Format	Choice		The format in which the generated images are returned Options: Base64 JSON, File Default: file
File Name	File	Yes	The configuration for the output file. The file extension will be determined by the output_type field
Output Type	Choice		The image file format for the output Options: PNG, JPEG Default: png
Retry Configuration	Object		Configure retry behavior for failed requests
Custom Tags	List		Optional custom tags for cost tracking (e.g., 'team:api', 'feat:summarize', 'env:prod')

Output

Later steps use this data by picking this step in the $ picker. For example, created is $.<step ID>.output.created.

{
  "created": 0,
  "data": [
    "string"
  ]
}

Example values show the shape of the data, not real values. See Using data from earlier steps.

Field descriptions

Field	Type	Description
created	Number	Unix timestamp (seconds) when images were generated
data[]	List	List of generated images

Analyze Image

Analyze an image (URL or base64) and return a text explanation

Settings

Field	Type	Required	Description
Model	Choice (loaded from your account)	Yes	Provider-prefixed model id from AI Gateway Default: openai/gpt-image-1.5
Image Source	Choice	Yes	The source of the image Options: URL, File
Image File	File	Yes	Image file to analyze (PNG, JPEG, WEBP, non-animated GIF)
Image URL	Text		HTTP/HTTPS URL for the image (PNG, JPEG, WEBP, non-animated GIF)
Prompt	Code	Yes	Instruction for how to analyze/describe the image
Retry Configuration	Object		Configure retry behavior for failed requests
Custom Tags	List		Optional custom tags for cost tracking (e.g., 'team:api', 'feat:summarize', 'env:prod')

Output

Later steps use this data by picking this step in the $ picker. For example, model_used is $.<step ID>.output.model_used.

{
  "model_used": "string",
  "raw_response": "string",
  "response": {
    "text": "string"
  },
  "success": false,
  "tokens_used": 0
}

Example values show the shape of the data, not real values. See Using data from earlier steps.

Field descriptions

Field	Type	Description
model_used	Text	Model used to perform image analysis
raw_response	Object	Response from the AI Gateway API
raw_response.created_at	Number	Unix timestamp when the response was created
raw_response.error	Text	Error message if the request failed
raw_response.id	Text	Response identifier
raw_response.incomplete_details	Object	Incomplete details if the response is incomplete
raw_response.instructions	Text	Instructions provided with the request
raw_response.max_output_tokens	Number	Maximum output tokens
raw_response.metadata	Object	Metadata object containing additional information
raw_response.model	Text	Model identifier used for the request
raw_response.object	Text	Object type (e.g., 'chat.completion')
raw_response.output[]	List	Output array containing the response items
raw_response.output[].content[]	List	Content array containing the response content
raw_response.output[].content[].annotations[]	List	Annotations array (can be empty). Elements are free-form annotation objects
raw_response.output[].content[].text	Text	Text content of the response
raw_response.output[].content[].type	Text	Content type (e.g., 'output_text')
raw_response.output[].id	Text	ID of the output item
raw_response.output[].role	Text	Role of the message (e.g., 'assistant')
raw_response.output[].status	Text	Status of the output item (e.g., 'completed')
raw_response.output[].type	Text	Type of output item (e.g., 'message')
raw_response.parallel_tool_calls	Yes / No	Whether parallel tool calls are enabled
raw_response.previous_response_id	Text	Previous response ID for continuation
raw_response.reasoning	Object	Reasoning object containing reasoning details
raw_response.status	Text	Status of the response (e.g., 'completed')
raw_response.store	Any	Store identifier (can be bool or string)
raw_response.temperature	Number	Temperature setting used for the request
raw_response.text	Object	Text object containing text-related information
raw_response.tool_choice	Text	Tool choice setting
raw_response.tools[]	List	Array of tools available for the request. Elements are free-form tool definitions
raw_response.top_p	Number	Top-p sampling parameter
raw_response.truncation	Text	Truncation information
raw_response.usage	Object	Usage statistics for the API request
raw_response.usage.cost	Any	Cost of the request
raw_response.usage.input_tokens	Number	Number of input tokens used
raw_response.usage.input_tokens_details	Any	Input tokens details
raw_response.usage.output_tokens	Number	Number of output tokens generated
raw_response.usage.output_tokens_details	Any	Output tokens details
raw_response.usage.total_tokens	Number	Total number of tokens used
raw_response.user	Text	User identifier
response	Object	Structured analysis response
response.text	Text	Textual explanation of the image
success	Yes / No	Whether the analyze_image request succeeded
tokens_used	Number	Tokens consumed when available

Transcribe Audio

Transcribe an attached or same-run reference audio file through AI Gateway

Settings

Field	Type	Required	Description
Model	Choice (loaded from your account)	Yes	Transcription model dynamically populated from AI Gateway Default: openai/whisper-1
Audio File	File	Yes	Audio file to transcribe (FLAC, MP3, MP4, MPEG, MPGA, M4A, OGG, WAV, or WEBM; max 25 MiB) Options: Reference, Attach new
Response Format	Choice		Format of the transcription response. Use 'diarized_json' for speaker separation with gpt-4o-transcribe-diarize. Options: JSON, Diarized JSON Default: json
Language	Text		Input audio language as ISO-639-1 code (e.g. en). ElevenLabs Scribe models.
Prompt	Text		Optional text to guide the model's style or vocabulary (e.g. proper nouns). Note: not supported for diarization models.
Temperature	Number		Sampling temperature between 0 and 1. ElevenLabs Scribe models.
Chunking Strategy	Choice		Chunking strategy for audio. Set to 'auto' for inputs longer than 30 seconds with gpt-4o-transcribe-diarize. Options: Auto
Custom Tags	List		Optional custom tags for AI Gateway cost tracking

Output

Later steps use this data by picking this step in the $ picker. For example, model_used is $.<step ID>.output.model_used.

{
  "model_used": "string",
  "segments": [
    {
      "end": 0,
      "id": "string",
      "speaker": "Speaker 1",
      "start": 0,
      "text": "string"
    }
  ],
  "text": "string"
}

Example values show the shape of the data, not real values. See Using data from earlier steps.

Field descriptions

Field	Type	Description
model_used	Text	AI Gateway model used for transcription
segments[]	List	Segments with timestamps and speaker labels when using diarized_json or verbose_json
segments[].end	Number	End timestamp in seconds
segments[].id	Any	Segment ID
segments[].speaker	Text	Speaker label (e.g. Speaker 1, Speaker 2)
segments[].start	Number	Start timestamp in seconds
segments[].text	Text	Transcribed segment text
text	Text	Transcribed text