Agents

The Agents section of SipPulse AI provides a powerful interface for creating and managing intelligent agents. These agents are customizable entities designed to perform specific functions, handle tasks, or simulate conversations based on their configuration and the instructions provided to them.

Getting Started with Agents

Creating an Agent

Initially, your agent list will be empty. To begin creating agents:

• Click the Create button to initiate a new agent.
• This action leads to a setup form where you can define the agent’s characteristics.

Unique Agent ID

Upon creation, each agent is assigned a unique ID, visible at the top left corner of the agent’s configuration page. This ID is crucial for integrating the agent with other parts of the platform, such as initiating threads or embedding the agent into applications. There is a convenient copy icon button next to the ID, allowing users to easily copy it for use in other parts of the system, such as creating threads.

Configuring Your Agent

After creation, you can customize your agent with several options to tailor its behavior for specific tasks:

• Name: Assign a name to your agent. This is just a label for easier identification and does not impact the agent's operational behavior.
• Instructions: Set directives or a persona for your agent, guiding how it should interact or respond in scenarios. Think of this as setting the tone and behavior of the agent.
• Model Selection: Choose from a variety of AI models. Some models come with a Tool tag, indicating enhanced capabilities that support advanced functions. Details on each model’s capabilities and suitable applications are provided to help you make an informed choice.
• Tools and Functions: You can augment your agent with specific tools or functions that extend its capabilities beyond basic interactions. This area allows for the integration of custom scripts or third-party APIs.
• Parameters: Adjust settings such as temperature and top-p to control the agent's response variability and creativity.

Agent Management

Each agent has buttons for effective management:

• Delete: Remove an agent permanently.
• Clone: Duplicate an existing agent. This is useful for creating similar agents without starting from scratch.

Testing and Interaction

To ensure your agent is set up correctly and to gauge its responses:

• Use the Playground feature by clicking the respective button. This sandbox environment lets you interact directly with your agent.
• During testing, you can see the agent's responses in real-time, which helps in fine-tuning its settings and instructions.

Metrics and Analytics in Playground

While interacting with your agent in the Playground, the interface provides valuable metrics:

• Cost Metrics: Understand the operational cost associated with using the agent.
• Token Consumption: Monitor how many tokens your interactions are consuming, helping manage usage and costs.
• Performance Metrics: View response times and other performance indicators to assess the efficiency of your agent.

This comprehensive setup allows for a high degree of customization and control, ensuring that agents created on SipPulse AI can effectively meet various operational needs. Whether for customer service, data retrieval, or interactive dialogues, these agents are equipped to handle complex interactions intelligently.

Tools and Enhancements for Agents

Agents on SipPulse AI can be enhanced with various tools to extend their capabilities. These tools are categorized into three main types: Functions, Files, and Built-in Tools. Each type serves different purposes and integrates different features to the agent.

Functions

The functions tool allows you to add and configure custom functions that your agent can call. This is an advanced feature that enhances the capabilities of your agent by enabling it to perform specific tasks programmatically.

Properties of a Function

1. Name: This is the identifier for your function. It should be written in snake_case. Although you can enter it in any format, it will ultimately be converted to snake_case.

2. Description: A detailed explanation of what the function does. This is crucial as it helps the agent understand when to call this function.

3. Parameters: This section uses JSON schema to define the structure of the parameters that the function will accept. JSON schema is a powerful tool for validating the structure of JSON data. Each parameter should include:

   • Type: Defines the data type (e.g., string, number, object).
   • Properties: Lists the individual properties if the parameter is an object.
   • Description: A brief explanation of what the parameter is used for.

   For more details on JSON schema, you can refer to the JSON Schema website.

Auto Call Feature

At the bottom of the function configuration, you will find the Auto Call feature.

• Auto Call Disabled: When this is turned off, the agent will invoke the function and wait for a user response to proceed.

• API Call Enabled: When this feature is enabled, the function will automatically call the API with the parameters defined by the user. The agent will:

  • Send a request to the specified API endpoint.
  • Interpret the response.
  • Continue the conversation based on the API's response.

  This is incredibly useful as it allows seamless integration of your own APIs into the agent's workflow. You can set up your API, define the parameters in the body or query, and the agent will handle the rest, ensuring a smooth and efficient conversation flow.

Example of Function Configuration

IMPORTANT

To reproduce the example, create and account in Amadeus for Developers and get an API key. Replace the url with your APP url and the MY_API_KEY with your actual key.

Below is an example of configuring a function to get flight information:

• Name: get_flight_info

• Description: Get flight information from a specific flight code.

• Parameters:

  {
    "type": "object",
    "properties": {
      "carrierCode": {
        "type": "string",
        "description": "Carrier code. Eg TP"
      },
      "flightNumber": {
        "type": "string",
        "description": "Flight number. Eg 1234"
      },
      "scheduledDepartureDate": {
        "type": "string",
        "description": "Scheduled departure date. Eg 2022-01-01"
      }
    },
    "required": ["carrierCode", "flightNumber", "scheduledDepartureDate"]
  }

• Auto Call: Enabled

  • API Call:
    • Method: GET
    • URL: https://my-app.api.amadeus.com/v2/schedule/flights
    • Headers:
      • Authorization
      • Bearer MY_API_KEY

This configuration allows the agent to automatically fetch flight information from the specified API when needed, providing a seamless user experience.

By setting up functions with detailed descriptions and parameters, and leveraging the Auto Call feature, you can significantly enhance the functionality of your agents, making them more interactive and capable of handling complex tasks autonomously.

Files

Files are a flexible tool where users can upload documents to assist the agent in generating responses based on the content of these documents. This feature is especially useful for agents requiring access to specific knowledge bases or reference materials.

• Name: A descriptive name for the file to help users remember what each file contains. This is purely for user reference.
• Description: A detailed description that the agent uses to determine the relevance of the file to a given conversation. This helps the agent understand what information is contained in the file and when to use it.

Document Handling Settings

To optimize the use of uploaded files, several settings can be configured:

• Embedding Model: This setting allows users to select the embedding model used to create vector representations of the document's content. Embeddings are crucial for enabling the agent to understand and retrieve relevant information from the document. While this process incurs a cost, it is significantly lower than typical model operations, making it a cost-effective way to enhance the agent's knowledge base.

• Document Segment Size: This setting defines the size of the chunks into which documents are split. The segmentation helps in handling and processing large documents more efficiently. Smaller segments allow the agent to retrieve more precise information relevant to a query, while larger segments may provide broader context. Adjusting the segment size can help balance detail and context based on the specific needs of the agent's tasks.

• Number of Documents Retrieved: This setting specifies how many document segments the agent should consider when generating responses. Limiting the number of retrieved segments can speed up response time and reduce processing costs, while retrieving more segments can improve the comprehensiveness of the agent's answers. Finding the right balance is key to optimizing performance and cost.

Built-in Tools

Built-in tools are pre-developed capabilities provided by SipPulse AI to aid agents in performing tasks that are complex for standard models. These tools are toggleable and can be enabled or disabled based on the needs of the agent. The current built-in tools include:

• Current Date and Time: Provides real-time clock information, enabling the agent to utilize the current date and time in conversations.
• Calculator: Allows the agent to perform basic arithmetic operations, extending its utility in mathematical contexts.
• Browser: Empowers the agent to perform web searches, fetching information directly from the internet to respond to queries effectively.

These tools collectively enhance an agent's ability to understand and interact based on a richer set of data and functionalities, making them more versatile and effective in varied application scenarios.