Integrating Large Language Models with External Tools: A Practical Guide to API Function Calls

In this tutorial we explore how we can connect OpenAI’s Chat Completions API to external services through function calling. This capability allows the model to generate JSON objects that can serve as instructions to call external functions based on user inputs.

AI gives you superpowers, it’s no longer just for chatting. Now, it can take real-world actions like sending emails, fetching live data, or updating databases. Explore how to set up structured functions, validate parameters, and handle responses to build dynamic AI applications that do more than just respond. Get ready to turn your chatbot into an active problem-solver that seamlessly connects with your entire tech setup. 

The Chat completion API provides us with an optional parameter “tools” that we can use to provide function definitions. The LLM will then select function arguments which match the provided function specifications.

One important thing to understand is that OpenAI doesn’t call any function. It simply returns the function definition, and the developer can then execute that function.

1. Optional “Tool” Parameter: The Chat Completion API allows you to use the optional “tool” parameter to provide function definitions for the LLM to reference.

2. LLM Function Selection: The LLM automatically selects function arguments that match the provided function specifications, allowing for dynamic function calling.

3. No Automatic Function Execution: OpenAI’s Chat Completion API does not execute the function. Instead, it returns the function definition to the developer.

4. Developer Responsibility: After receiving the function definition, it’s the developer's responsibility to execute the function with the selected arguments, integrating it into the application. 

The completion response is computed and after that it is returned.

1. Make calls to external APIs: Use tools to make calls to external api. Build agents that can understand and perform external tasks by defining functions. For example: generating user location or fetching pending tasks for this week:

get_pending_tasks(type: ‘week|month|quarter’)
get_current_location() 
send_friend_request(userId) · 

2. Connect to SQL Database: Using Natural language, define functions that can make request to database queries.

get_data_from_db(query)

Integrating LLMs with external tools involves a few straightforward steps:  

1. Define Functions: When making an API call, specify the functions that the LLM can choose to call. Specify the function definitions in the “tools” parameter.  

Define Functions

2. Call the Model: Send the user’s query to the model. The model processes the input and determines whether and which function to call, returning a JSON object that represents the function call.

3. Handle the Output: The JSON output from the model will contain the necessary details to call the external function. This may include function names and argument values.

If a tool was called, the “finish_reason“ in the ocmpletion respnse will be equal to “tool_calls“. Also, there will be a “tool_calls” object that will contain the name of tool as well as any arguments.

4. Execute Functions: Parse the JSON and execute the function calls in your system as required.

5. Process and Respond: After executing the function, you might need to call the model again with the results to generate a user-friendly summary or further instructions.

Integrating OpenAI’s Chat Completions API with external tools improves the functional capabilities of applications. By defining functions and utilizing the ‘tool’ parameter, developers can create interactions that use real-time data and actions to increase the responsiveness and versatility of their software.

Ready to improve your business operations by innovating Data into conversations? Click here to see how Data Outlook can help you automate your processes.