Skip to main content

Tools (Function Calling)

Some LLMs, in addition to generating text, can also trigger actions.

There is a concept known as "tools," or "function calling". It allows the LLM to call, when necessary, one or more available tools, usually defined by the developer. A tool can be anything: a web search, a call to an external API, or the execution of a specific piece of code, etc. LLMs cannot actually call the tool themselves; instead, they express the intent to call a specific tool in their response (instead of responding in plain text). We, as developers, should then execute this tool with the provided arguments and report back the results of the tool execution.

For example, we know that LLMs themselves are not very good at math. If your use case involves occasional math calculations, you might want to provide the LLM with a "math tool." By declaring one or multiple tools in the request to the LLM, it can then decide to call one of them if it deems it appropriate. Given a math question along with a set of math tools, the LLM might decide that to properly answer the question, it should first call one of the provided math tools.

Let's see how this works in practice (with and without tools):

An example of a message exchange without tools:

- messages:
- UserMessage:
- text: What is the square root of 475695037565?

- AiMessage:
- text: The square root of 475695037565 is approximately 689710.

Close, but not correct.

An example of a message exchange with the following tools:

@Tool("Sums 2 given numbers")
public double sum(double a, double b) {
return a + b;

@Tool("Returns a square root of a given number")
public double squareRoot(double x) {
return Math.sqrt(x);
Request 1:
- messages:
- UserMessage:
- text: What is the square root of 475695037565?
- tools:
- sum(double a, double b): Sums 2 given numbers
- squareRoot(double x): Returns a square root of a given number

Response 1:
- AiMessage:
- toolExecutionRequests:
- squareRoot(475695037565)

... here we are executing the squareRoot method with the "475695037565" argument and getting "689706.486532" as a result ...

Request 2:
- messages:
- UserMessage:
- text: What is the square root of 475695037565?
- AiMessage:
- toolExecutionRequests:
- squareRoot(475695037565)
- ToolExecutionResultMessage:
- text: 689706.486532

Response 2:
- AiMessage:
- text: The square root of 475695037565 is 689706.486532.

As you can see, when an LLM has access to tools, it can decide to call one of them when appropriate.

This is a very powerful feature. In this simple example, we gave the LLM primitive math tools, but imagine if we gave it, for example, googleSearch and sendEmail tools and a query like "My friend wants to know recent news in the AI field. Send the short summary to," then it could find recent news using the googleSearch tool, then summarize it and send the summary via email using the sendEmail tool.


To increase the chances of the LLM calling the right tool with the right arguments, we should provide a clear and unambiguous:

  • name of the tool
  • description of what the tool does and when it should be used
  • description of every tool parameter

A good rule of thumb: if a human can understand the purpose of a tool and how to use it, chances are that the LLM can too.

LLMs are specifically fine-tuned to detect when to call tools and how to call them. Some models can even call multiple tools at once, for example, OpenAI.


Please note that not all models support tools. Currently, the following models have tool support:

  • OpenAiChatModel
  • AzureOpenAiChatModel
  • LocalAiChatModel
  • QianfanChatModel

Please note that tools/function calling is not the same as JSON mode.

2 levels of abstraction

LangChain4j provides two levels of abstraction for using tools.

Low level Tool API

At the low level, you can use the generate(List<ChatMessage>, List<ToolSpecification>) method of the ChatLanguageModel. A similar method is also present in the StreamingChatLanguageModel.

ToolSpecification is an object that contains all the information about the tool:

  • The name of the tool
  • The description of the tool
  • The parameters (arguments) of the tool and their descriptions

It is recommended to provide as much information about the tool as possible: a clear name, a comprehensive description, and a description for each parameter, etc.

There are two ways to create a ToolSpecification:

  1. Manually
ToolSpecification toolSpecification = ToolSpecification.builder()
.description("Returns the weather forecast for a given city")
.addParameter("city", type("string"), description("The city for which the weather forecast should be returned"))
.addParameter("temperatureUnit", enums(TemperatureUnit.class)) // enum TemperatureUnit { CELSIUS, FAHRENHEIT }
  1. Using helper methods:
  • ToolSpecifications.toolSpecificationsFrom(Class)
  • ToolSpecifications.toolSpecificationsFrom(Object)
  • ToolSpecifications.toolSpecificationFrom(Method)
class WeatherTools { 

@Tool("Returns the weather forecast for a given city")
String getWeather(
@P("The city for which the weather forecast should be returned") String city,
TemperatureUnit temperatureUnit
) {

List<ToolSpecification> toolSpecifications = ToolSpecifications.toolSpecificationsFrom(WeatherTools.class);

Once you have a List<ToolSpecification>, you can call the model:

UserMessage userMessage = UserMessage.from("What will the weather be like in London tomorrow?");
Response<AiMessage> response = model.generate(singletoneList(userMessage), toolSpecifications);
AiMessage aiMessage = response.content();

If the LLM decides to call the tool, the returned AiMessage will contain data in the toolExecutionRequests field. In this case, AiMessage.hasToolExecutionRequests() will return true. Depending on the LLM, it can contain one or multiple ToolExecutionRequest objects (some LLMs support calling multiple tools in parallel).

Each ToolExecutionRequest should contain:

  • The id of the tool call (some LLMs do not provide it)
  • The name of the tool to be called, for example: getWeather
  • The arguments, for example: { "city": "London", "temperatureUnit": "CELSIUS" }

You'll need to manually execute the tool(s) using information from the ToolExecutionRequest(s).

If you want to send the result of the tool execution back to the LLM, you need to create a ToolExecutionResultMessage (one for each ToolExecutionRequest) and send it along with all previous messages:

String result = "It is expected to rain in London tomorrow.";
ToolExecutionResultMessage toolExecutionResultMessage = ToolExecutionResultMessage.from(toolExecutionRequest, result);
List<ChatMessage> messages = List.of(userMessage, aiMessage, toolExecutionResultMessage);
Response<AiMessage> response2 = model.generate(messages, toolSpecifications);

High Level Tool API

At a high level, you can annotate any Java method with the @Tool annotation and use it with AI Services.

AI Services will automatically convert such methods into ToolSpecifications and include them in the request for each interaction with the LLM. When the LLM decides to call the tool, the AI Service will automatically execute the appropriate method, and the return value of the method (if any) will be sent back to the LLM. You can find implementation details in DefaultToolExecutor.

Methods annotated with @Tool can accept any number of parameters of various types.

They can also return any type, including void. If the method has a void return type, "Success" string is sent to the LLM if the method returns successfully.

If the method has a String return type, the returned value is sent to the LLM as is, without any conversions.

For other return types, the returned value is converted into a JSON before being sent to the LLM.

A few tool examples:

@Tool("Searches Google for relevant URLs, given the query")
public List<String> searchGoogle(@P("search query") String query) {

@Tool("Returns the content of a web page, given the URL")
public String getWebPageContent(@P("URL of the page") String url) {
Document jsoupDocument = Jsoup.connect(url).get();
return jsoupDocument.body().text();


Any Java method annotated with @Tool and explicitly specified during the build of an AI Service can be executed by the LLM:

interface MathGenius {

String ask(String question);

class Calculator {

public double add(int a, int b) {
return a + b;

public double squareRoot(double x) {
return Math.sqrt(x);

MathGenius mathGenius = AiServices.builder(MathGenius.class)
.tools(new Calculator())

String answer = mathGenius.ask("What is the square root of 475695037565?");

System.out.println(answer); // The square root of 475695037565 is 689706.486532.

When the ask method is called, 2 interactions with the LLM occur, as described in the earlier section. In between those interactions, the squareRoot method is called automatically.

The @Tool annotation has 2 optional fields:

  • name: the tool's name. If this is not provided, the method's name will serve as the tool's name.
  • value: the tool's description.

Depending on the tool, the LLM might understand it well even without any description (for example, add(a, b) is obvious), but it is usually better to provide clear and meaningful names and descriptions. This way, the LLM has more information to decide whether or not to call the given tool, and how to do so.


Method parameters can optionally be annotated with @P. @P has a single mandatory field (value) for providing a description of the parameter.


If your AI Service method has a parameter annotated with @MemoryId, you can also annotate a parameter of a @Tool method with @ToolMemoryId. The value provided to the AI Service method will be automatically passed to the @Tool method. This feature is useful if you have multiple users and/or multiple chats/memories per user and wish to distinguish between them inside the @Tool method.