Generic AI Agent

Inputs

Name	Type	Description
Model Name	String	The LLM model to use
System Prompt	String	The system prompt for the agent
Conversation	Object	The conversation history (Conversation object or list of OpenAI-style messages)
Force Tool Usage	Optional Boolean	When enabled, the agent only produces function calls. Defaults to `false`
Tools	Optional List	List of tool definitions
Response Format	Optional Object	Response format specification. Defaults to null
Temperature	Optional Number	Measure of randomness in generated text. Defaults to `0`

Model options: gpt-5, gpt-5-mini, gpt-4o, gpt-4o-mini, gpt-4.1, azure-gpt-5, azure-gpt-5-mini, azure-gpt-4o, azure-gpt-4o-mini, azure-gpt-41, gemini-2.0-flash, gemini-2.5-pro, gemini-2.5-flash

Outputs

Name	Type	Description
Content	String	The response content from the agent
Tool Call	String	The tool call name if a tool was called
Arguments	Object	The arguments for the tool call
Success	Boolean	Whether the agent execution was successful
Server Response	Object	The server response (e.g. on error)

OpenAI-style messages example

Conversation can be a list of messages in OpenAI format: each item has role and content.

[
  {"role": "user", "content": "Hi, I need help with a return."},
  {"role": "assistant", "content": "I'd be happy to help you with your return."},
  {"role": "user", "content": "I bought shoes that don't fit. Can I return them?"}
]

Tools example

Tools is a list of tool definitions. Each tool has a name, description, and parameters (JSON schema). The agent can choose to call one of these tools; when it does, the tool name is returned in Tool Call and the arguments in Arguments.

Simple tool (no parameters):

[
  {
    "name": "order_status",
    "description": "Use this tool to check on customer's order status.",
    "parameters": {
      "properties": {},
      "required": []
    }
  }
]

Tool with required parameters:

{
  "name": "information_collector",
  "description": "This tool is responsible for collecting information",
  "parameters": {
    "type": "object",
    "properties": {
      "order_number": {
        "type": "string",
        "description": "Order number containing 7 digits"
      }
    },
    "required": ["order_number"]
  }
}

Response format example

Response Format constrains the agent output to a JSON structure. When specified, the model returns structured text that fits the schema. That text is returned as a string in Content.

{
  "type": "json_schema",
  "json_schema": {
    "name": "ConversationEvaluation",
    "description": "An evaluation of an E-commerce customer service conversation for a given criteria.",
    "schema": {
      "type": "object",
      "properties": {
        "explanation": {
          "description": "The explanation for the verdict, based on the rubric and conversation evidence.",
          "type": "string"
        },
        "score": {
          "description": "A score for the criterion on a 0.0-1.0 scale (in 0.1 increments)",
          "type": "number"
        }
      },
      "required": ["explanation", "score"],
      "additionalProperties": false
    }
  }
}

Notes

On success, the action returns Content (and optionally Tool Call + Arguments when a tool was called) with success: true.
If Response Format is specified, the structured output is returned as a string in Content. Parse it into the corresponding type (e.g. JSON).
When the agent calls a tool, Tool Call is the tool name and Arguments holds the arguments for that call.
On failure, the action returns success: false and may include Server Response with error details.