Overview

Integry Functions enables you to integrate a vast array of apps into your AI applications seamlessly.

What are Integry Functions?

Integry Functions lets you use Integry's extensive library of app functions as tools within your AI applications. By leveraging Integry's function-calling compatible JSON structures, you can invoke third-party app functions directly from your AI calls. You can:

Access hundreds of third-party app functions.
Use function-calling compatible JSON structures in your AI calls.
Show a populated form of the action for user editing and confirmation
Invoke functions directly (headlessly i.e. without a UI)
Predict and populate functions based on user input.

Integry supports many functions for 200+ apps. Click the Marketplace button in the app to browse the full list of Functions or use the API.

Using Functions with Your AI

Generally, when using AI with function calling, you take the following steps:

Collect User Input: You collect a user message, which requires talking to an external API. For example: "send hello to my team on slack on #general"
Call AI Model: You take the user message, their data context, and a prompt and pass a list of function specs with to the AI's API call
AI Populates Function: The AI selects the function, fills in its parameters, and returns the result
Call Function: You then call the function with the arguments and take next steps based on the response
Validate Arguments (optional): before calling the function in step 3, you might need to validate the user input. In our example, the user wrote #general whereas the actual channel was #general. To resolve this, you may need to first fetch all channels and find the closest match
Show UI (optional): you could also show the user the final function call in a UI before executing it.

To achieve this, we have two methods. The first method is faster and easier to start with. The second method requires more effort but provides greater control. Here they are:

Function Prediction: Pass the user message to/functions/?predict_function=true&prompt=<user_message> to get the most relevant function. You can also get populated arguments.
Function List Retrieval: Retrieve a list of available functions using the /functions endpoint and pass them to your AI model. Your model will predict the function to use and will populate the arguments. You may need to validate your inputs with additional supporting function calls.

Both approaches converge after obtaining the function parameters, where you can choose to call the function directly or show the UI.

Additional Resources

PreviousZoom NextViewing Function Calls

Last updated 1 month ago

What are Integry Functions?

Access hundreds of third-party app functions.

Use function-calling compatible JSON structures in your AI calls.

Show a populated form of the action for user editing and confirmation

Invoke functions directly (headlessly i.e. without a UI)

Predict and populate functions based on user input.

Integry supports many functions for 200+ apps. Click the Marketplace button in the app to browse the full list of Functions or use the API.

Using Functions with Your AI

Generally, when using AI with function calling, you take the following steps:

Collect User Input: You collect a user message, which requires talking to an external API. For example: "send hello to my team on slack on #general"

Call AI Model: You take the user message, their data context, and a prompt and pass a list of function specs with to the AI's API call

AI Populates Function: The AI selects the function, fills in its parameters, and returns the result

Call Function: You then call the function with the arguments and take next steps based on the response

Validate Arguments (optional): before calling the function in step 3, you might need to validate the user input. In our example, the user wrote #general whereas the actual channel was #general. To resolve this, you may need to first fetch all channels and find the closest match

Show UI (optional): you could also show the user the final function call in a UI before executing it.

To achieve this, we have two methods. The first method is faster and easier to start with. The second method requires more effort but provides greater control. Here they are:

Function Prediction: Pass the user message to/functions/?predict_function=true&prompt=<user_message> to get the most relevant function. You can also get populated arguments.

Function List Retrieval: Retrieve a list of available functions using the /functions endpoint and pass them to your AI model. Your model will predict the function to use and will populate the arguments. You may need to validate your inputs with additional supporting function calls.

Both approaches converge after obtaining the function parameters, where you can choose to call the function directly or show the UI.