Skip to content

Latest commit

 

History

History
112 lines (88 loc) · 3.63 KB

agent_functions.md

File metadata and controls

112 lines (88 loc) · 3.63 KB

📘 Delphi Tutorial: Using Agent Functions

In Delphi, agent functions are a versatile tool allowing your conversational agent to execute specific tasks, like data retrieval or complex computations, dynamically during a conversation. Let's dive into how you can utilize these functions in your Delphi projects.

Defining Functions

  • Custom Functions: You can define custom functions tailored to your specific needs. These functions can perform various tasks like fetching data from an API, processing input, or carrying out computations.
  • Schema Validation: Delphi uses JSON schemas to validate the inputs of these functions, ensuring data integrity and correctness.

Registering Functions

  • Function Registration: After defining a function, you need to register it with the context. This registration process tells the context which functions are available for use during a conversation.

    const myFunction = new AgentFunction(/* ... configuration ... */);
context.addFunction(myFunction, true); // second argument is whether to enable immediately

    For more information on the AgentFunction class, see the API reference.

Enabling And Disabling Functions

  • Enabling: Once registered, a function may not be immediately enabled. This step offers control over which functions are active at any given time. To enable a function, use the enable method on the functions property of the context.

    context.functions.enable(myFunction); // or by name

  • Disabling: To disable a function, use the disable method on the functions property of the context.

    context.functions.disable(myFunction); // or by name

    Disabled functions will not be available to the agent during conversations.

Function Invocation

  • During Conversations: The model may invoke functions during a conversation. This will enhance the agent's context, allowing it to respond more intelligently to the user, and to perform more complex tasks.

Handling Responses

  • Response Integration: The results from function calls are seamlessly integrated into the conversation. Depending on your function's design, it can directly influence the agent's next response or modify the conversation's context.

Example

import { AgentFunction, type JSONSchemaType } from "@wecandobetter/delphi";

interface WeatherParameters {
  location: string;
}

interface WeatherData {
  temperature: number;
  humidity: number;
  windSpeed: number;
}

const weatherSchema: JSONSchemaType<WeatherParameters> = {
  type: "object",
  properties: {
    location: {
      type: "string",
      description:
        "The location to fetch weather information for (e.g. 'The Hague, NL')",
    },
  },
  required: ["location"],
  additionalProperties: false,
};

// Define a custom function
const fetchWeatherInfo = new AgentFunction<WeatherParameters, WeatherData>(
  "getWeather",
  "Fetch current weather information",
  weatherSchema, // Your defined JSON schema for input validation
  async (params) => {
    // Logic to fetch weather information
    return {
      temperature: 22.5, // in Celsius
      humidity: 0.5,
      windSpeed: 10,
  },
);

Conclusion

Agent functions in Delphi provide a powerful way to extend the capabilities of your conversational agents. By defining custom functions, registering them with your agent, and then utilizing them in conversation flows, you can create rich, interactive, and dynamic conversational experiences. The key is to tailor these functions to fit the specific needs and contexts of your application. Happy coding!