MistralAI
Project setup
To install langchain4j to your project, add the following dependency:
For Maven project pom.xml
<dependency>
<groupId>dev.langchain4j</groupId>
<artifactId>langchain4j</artifactId>
<version>0.35.0</version>
</dependency>
<dependency>
<groupId>dev.langchain4j</groupId>
<artifactId>langchain4j-mistral-ai</artifactId>
<version>0.35.0</version>
</dependency>
For Gradle project build.gradle
implementation 'dev.langchain4j:langchain4j:0.35.0'
implementation 'dev.langchain4j:langchain4j-mistral-ai:0.35.0'
API Key setup
Add your MistralAI API key to your project, you can create a class ApiKeys.java
with the following code
public class ApiKeys {
public static final String MISTRALAI_API_KEY = System.getenv("MISTRAL_AI_API_KEY");
}
Don't forget set your API key as an environment variable.
export MISTRAL_AI_API_KEY=your-api-key #For Unix OS based
SET MISTRAL_AI_API_KEY=your-api-key #For Windows OS
More details on how to get your MistralAI API key can be found here
Model Selection
You can use MistralAiChatModelName.class
enum class to found appropriate model names for your use case.
MistralAI updated a new selection and classification of models according to performance and cost trade-offs.
Model name | Deployment or available from | Description |
---|---|---|
open-mistral-7b | - Mistral AI La Plateforme. - Cloud platforms (Azure, AWS, GCP). - Hugging Face. - Self-hosted (On-premise, IaaS, docker, local). | OpenSource The first dense model released by Mistral AI, perfect for experimentation, customization, and quick iteration. Max tokens 32K Java Enum MistralAiChatModelName.OPEN_MISTRAL_7B |
open-mixtral-8x7b | - Mistral AI La Plateforme. - Cloud platforms (Azure, AWS, GCP). - Hugging Face. - Self-hosted (On-premise, IaaS, docker, local). | OpenSource Ideal to handle multi-languages operations, code generationand fine-tuned. Excellent cost/performance trade-offs. Max tokens 32K Java Enum MistralAiChatModelName.OPEN_MIXTRAL_8x7B |
open-mixtral-8x22b | - Mistral AI La Plateforme. - Cloud platforms (Azure, AWS, GCP). - Hugging Face. - Self-hosted (On-premise, IaaS, docker, local). | OpenSource It has all Mixtral-8x7B capabilities plus strong maths and coding natively capable of function calling Max tokens 64K. Java Enum MistralAiChatModelName.OPEN_MIXTRAL_8X22B |
mistral-small-latest | - Mistral AI La Plateforme. - Cloud platforms (Azure, AWS, GCP). | Commercial Suitable for simple tasks that one can do in bulk (Classification, Customer Support, or Text Generation). Max tokens 32K Java Enum MistralAiChatModelName.MISTRAL_SMALL_LATEST |
mistral-medium-latest | - Mistral AI La Plateforme. - Cloud platforms (Azure, AWS, GCP). | Commercial Ideal for intermediate tasks that require moderate reasoning (Data extraction, Summarizing, Writing emails, Writing descriptions. Max tokens 32K Java Enum MistralAiChatModelName.MISTRAL_MEDIUM_LATEST |
mistral-large-latest | - Mistral AI La Plateforme. - Cloud platforms (Azure, AWS, GCP). | Commercial Ideal for complex tasks that require large reasoning capabilities or are highly specialized (Text Generation, Code Generation, RAG, or Agents). Max tokens 32K Java Enum MistralAiChatModelName.MISTRAL_LARGE_LATEST |
mistral-embed | - Mistral AI La Plateforme. - Cloud platforms (Azure, AWS, GCP). | Commercial Converts text into numerical vectors of embeddings in 1024 dimensions. Embedding models enable retrieval and RAG applications. Max tokens 8K Java Enum MistralAiEmbeddingModelName.MISTRAL_EMBED |
@Deprecated
models:
- mistral-tiny (
@Deprecated
) - mistral-small (
@Deprecated
) - mistral-medium (
@Deprecated
)
You can find more detail and types of use cases with their respective Mistral model here
Chat Completion
The chat models allow you to generate human-like responses with a model fined-tuned on conversational data.
Synchronous
Create a class and add the following code.
import dev.langchain4j.model.chat.ChatLanguageModel;
import dev.langchain4j.model.mistralai.MistralAiChatModel;
public class HelloWorld {
public static void main(String[] args) {
ChatLanguageModel model = MistralAiChatModel.builder()
.apiKey(ApiKeys.MISTRALAI_API_KEY)
.modelName(MistralAiChatModelName.MISTRAL_SMALL_LATEST)
.build();
String response = model.generate("Say 'Hello World'");
System.out.println(response);
}
}
Running the program will generate a variant of the following output
Hello World! How can I assist you today?
Streaming
Create a class and add the following code.
import dev.langchain4j.data.message.AiMessage;
import dev.langchain4j.model.StreamingResponseHandler;
import dev.langchain4j.model.mistralai.MistralAiStreamingChatModel;
import dev.langchain4j.model.output.Response;
import java.util.concurrent.CompletableFuture;
public class HelloWorld {
public static void main(String[] args) {
MistralAiStreamingChatModel model = MistralAiStreamingChatModel.builder()
.apiKey(ApiKeys.MISTRALAI_API_KEY)
.modelName(MistralAiChatModelName.MISTRAL_SMALL_LATEST)
.build();
CompletableFuture<Response<AiMessage>> futureResponse = new CompletableFuture<>();
model.generate("Tell me a joke about Java", new StreamingResponseHandler() {
@Override
public void onNext(String token) {
System.out.print(token);
}
@Override
public void onComplete(Response<AiMessage> response) {
futureResponse.complete(response);
}
@Override
public void onError(Throwable error) {
futureResponse.completeExceptionally(error);
}
});
futureResponse.join();
}
}
You will receive each chunk of text (token) as it is generated by the LLM on the onNext
method.
You can see that output below is streamed in real-time.
"Why do Java developers wear glasses? Because they can't C#"
Of course, you can combine MistralAI chat completion with other features like Set Model Parameters and Chat Memory to get more accurate responses.
In Chat Memory you will learn how to pass along your chat history, so the LLM knows what has been said before. If you don't pass the chat history, like in this simple example, the LLM will not know what has been said before, so it won't be able to correctly answer the second question ('What did I just ask?').
A lot of parameters are set behind the scenes, such as timeout, model type and model parameters. In Set Model Parameters you will learn how to set these parameters explicitly.
Function Calling
Function calling allows Mistral chat models (synchronous and streaming) to connect to external tools. For example, you can call a Tool
to get the payment transaction status as shown in the Mistral AI function calling tutorial.
What are the supported mistral models?
Currently, function calling is available for the following models:
- Mistral Small
MistralAiChatModelName.MISTRAL_SMALL_LATEST
- Mistral Large
MistralAiChatModelName.MISTRAL_LARGE_LATEST
- Mixtral 8x22B
MistralAiChatModelName.OPEN_MIXTRAL_8X22B
1. Define a Tool
class and how get the payment data
Let's assume you have a dataset of payment transaction like this. In real applications you should inject a database source or REST API client to get the data.
import java.util.*;
public class PaymentTransactionTool {
private final Map<String, List<String>> paymentData = Map.of(
"transaction_id", List.of("T1001", "T1002", "T1003", "T1004", "T1005"),
"customer_id", List.of("C001", "C002", "C003", "C002", "C001"),
"payment_amount", List.of("125.50", "89.99", "120.00", "54.30", "210.20"),
"payment_date", List.of("2021-10-05", "2021-10-06", "2021-10-07", "2021-10-05", "2021-10-08"),
"payment_status", List.of("Paid", "Unpaid", "Paid", "Paid", "Pending"));
...
}
Next, let's define two methods retrievePaymentStatus
and retrievePaymentDate
to get the payment status and payment date from the Tool
class.
// Tool to be executed to get payment status
@Tool("Get payment status of a transaction") // function description
String retrievePaymentStatus(@P("Transaction id to search payment data") String transactionId) {
return getPaymentData(transactionId, "payment_status");
}
// Tool to be executed to get payment date
@Tool("Get payment date of a transaction") // function description
String retrievePaymentDate(@P("Transaction id to search payment data") String transactionId) {
return getPaymentData(transactionId, "payment_date");
}
private String getPaymentData(String transactionId, String data) {
List<String> transactionIds = paymentData.get("transaction_id");
List<String> paymentData = paymentData.get(data);
int index = transactionIds.indexOf(transactionId);
if (index != -1) {
return paymentData.get(index);
} else {
return "Transaction ID not found";
}
}
It uses a @Tool
annotation to define the function description and @P
annotation to define the parameter description of the package dev.langchain4j.agent.tool.*
. More info here
2. Define an interface as an agent
to send chat messages.
Create an interface PaymentTransactionAgent
.
import dev.langchain4j.service.SystemMessage;
interface PaymentTransactionAgent {
@SystemMessage({
"You are a payment transaction support agent.",
"You MUST use the payment transaction tool to search the payment transaction data.",
"If there a date convert it in a human readable format."
})
String chat(String userMessage);
}
3. Define a main
application class to chat with the MistralAI chat model
import dev.langchain4j.memory.chat.MessageWindowChatMemory;
import dev.langchain4j.model.chat.ChatLanguageModel;
import dev.langchain4j.model.mistralai.MistralAiChatModel;
import dev.langchain4j.model.mistralai.MistralAiChatModelName;
import dev.langchain4j.service.AiServices;
public class PaymentDataAssistantApp {
ChatLanguageModel mistralAiModel = MistralAiChatModel.builder()
.apiKey(System.getenv("MISTRAL_AI_API_KEY")) // Please use your own Mistral AI API key
.modelName(MistralAiChatModelName.MISTRAL_LARGE_LATEST) // Also you can use MistralAiChatModelName.OPEN_MIXTRAL_8X22B as open source model
.logRequests(true)
.logResponses(true)
.build();
public static void main(String[] args) {
// STEP 1: User specify tools and query
PaymentTransactionTool paymentTool = new PaymentTransactionTool();
String userMessage = "What is the status and the payment date of transaction T1005?";
// STEP 2: User asks the agent and AiServices call to the functions
PaymentTransactionAgent agent = AiServices.builder(PaymentTransactionAgent.class)
.chatLanguageModel(mistralAiModel)
.tools(paymentTool)
.chatMemory(MessageWindowChatMemory.withMaxMessages(10))
.build();
// STEP 3: User gets the final response from the agent
String answer = agent.chat(userMessage);
System.out.println(answer);
}
}
and expect an answer like this:
The status of transaction T1005 is Pending. The payment date is October 8, 2021.
JSON mode
You can also use the JSON mode to get the response in JSON format. To do this, you need to set the responseFormat
parameter to json_object
or the java enum MistralAiResponseFormatType.JSON_OBJECT
in the MistralAiChatModel
builder OR MistralAiStreamingChatModel
builder.
Syncronous example:
ChatLanguageModel model = MistralAiChatModel.builder()
.apiKey(System.getenv("MISTRAL_AI_API_KEY")) // Please use your own Mistral AI API key
.responseFormat(MistralAiResponseFormatType.JSON_OBJECT)
.build();
String userMessage = "Return JSON with two fields: transactionId and status with the values T123 and paid.";
String json = model.generate(userMessage);
System.out.println(json); // {"transactionId":"T123","status":"paid"}
Streaming example:
StreamingChatLanguageModel streamingModel = MistralAiStreamingChatModel.builder()
.apiKey(System.getenv("MISTRAL_AI_API_KEY")) // Please use your own Mistral AI API key
.responseFormat(MistralAiResponseFormatType.JSON_OBJECT)
.build();
String userMessage = "Return JSON with two fields: transactionId and status with the values T123 and paid.";
CompletableFuture<Response<AiMessage>> futureResponse = new CompletableFuture<>();
streamingModel.generate(userMessage, new StreamingResponseHandler() {
@Override
public void onNext(String token) {
System.out.print(token);
}
@Override
public void onComplete(Response<AiMessage> response) {
futureResponse.complete(response);
}
@Override
public void onError(Throwable error) {
futureResponse.completeExceptionally(error);
}
});
String json = futureResponse.get().content().text();
System.out.println(json); // {"transactionId":"T123","status":"paid"}
Guardrailing
Guardrails are a way to limit the behavior of the model to prevent it from generating harmful or unwanted content. You can set optionally safePrompt
parameter in the MistralAiChatModel
builder or MistralAiStreamingChatModel
builder.
Syncronous example:
ChatLanguageModel model = MistralAiChatModel.builder()
.apiKey(System.getenv("MISTRAL_AI_API_KEY"))
.safePrompt(true)
.build();
String userMessage = "What is the best French cheese?";
String response = model.generate(userMessage);
Streaming example:
StreamingChatLanguageModel streamingModel = MistralAiStreamingChatModel.builder()
.apiKey(System.getenv("MISTRAL_AI_API_KEY"))
.safePrompt(true)
.build();
String userMessage = "What is the best French cheese?";
CompletableFuture<Response<AiMessage>> futureResponse = new CompletableFuture<>();
streamingModel.generate(userMessage, new StreamingResponseHandler() {
@Override
public void onNext(String token) {
System.out.print(token);
}
@Override
public void onComplete(Response<AiMessage> response) {
futureResponse.complete(response);
}
@Override
public void onError(Throwable error) {
futureResponse.completeExceptionally(error);
}
});
futureResponse.join();
Toggling the safe prompt will prepend your messages with the following @SystemMessage
:
Always assist with care, respect, and truth. Respond with utmost utility yet securely. Avoid harmful, unethical, prejudiced, or negative content. Ensure replies promote fairness and positivity.