OpenAI

This page outlines how to use the OpenAI integration for AI-powered actions in Cargo.

OpenAI provides advanced language models that power intelligent automation. Cargo’s native integration with OpenAI allows you to use GPT models directly in your workflows for text generation, classification, summarization, and more.

How to set up OpenAI

You can use OpenAI in two ways:

1. Cargo credits – Use OpenAI through Cargo’s managed integration (no API key required)
2. Your own API key – Connect your OpenAI account for direct access

Using Cargo credits

Select OpenAI from the node catalog and start using it immediately. Costs are deducted from your Cargo credits based on token usage.

Using your own API key

To connect your OpenAI account:

Field	Description
API Key	Your OpenAI API key

Find your API key in the OpenAI platform under API keys.

---

OpenAI actions

Instruct

Send a prompt to an OpenAI model and receive a text response.

Use cases

• Content generation – Generate personalized emails, messages, or content
• Classification – Categorize text into predefined categories
• Summarization – Summarize long text into concise summaries
• Data extraction – Extract structured data from unstructured text
• Translation – Translate text between languages
• Analysis – Analyze sentiment, intent, or other text properties

Configuration

Field	Description
Model	Select the GPT model to use
Prompt	The prompt to send to the model
System prompt	Optional system instructions
Maximum tokens	Limit output length (includes reasoning tokens)
Temperature	Control randomness (0 = deterministic, 2 = very random)
With web search	Enable web search capabilities for the agent
Response format	Text, JSON object, or JSON schema

Available models

The available models — and their per-model credit costs — evolve with OpenAI’s lineup, so the action’s Model dropdown in the workspace is always the authoritative list. From the terminal:

cargo-ai connection integration get openAi

Picking a model:

• Flagship models — complex reasoning, multi-step analysis, coding tasks
• Mini variants — the best cost/performance balance for most everyday tasks (start here)
• Nano variants — high-volume, low-stakes work: summarization, classification, simple extraction

Response formats

Control how the model returns its response:

Format	Description
Text	Free-form text response (default)
JSON object	Response formatted as valid JSON
JSON schema	Response validated against a specific JSON schema

Use JSON schema when you need structured output with specific fields. Define your schema to ensure consistent, parseable responses.

Example JSON schema

{
  "type": "object",
  "properties": {
    "sentiment": { "type": "string", "enum": ["positive", "negative", "neutral"] },
    "confidence": { "type": "number" },
    "summary": { "type": "string" }
  },
  "additionalProperties": false,
  "required": ["sentiment", "confidence", "summary"]
}

Advanced settings

System prompt

Set instructions that guide the model’s behavior across all prompts. Use system prompts to:

• Define the model’s role or persona
• Set output format guidelines
• Provide context about your use case
• Establish constraints or rules

Temperature

Control the randomness of outputs:

Temperature	Behavior
0	Most deterministic, consistent outputs
0.5-1	Balanced creativity and consistency
1-2	More creative, varied outputs

For tasks requiring consistency (classification, extraction), use low temperature (0-0.3). For creative tasks (content generation), use higher temperature (0.7-1).

Web search

Enable the model to search the web for current information. Useful for:

• Real-time data lookups
• Current events or news
• Company or product research
• Fact verification