diff --git a/examples/12_function_calling/function_calling.go b/examples/12_function_calling/function_calling.go
new file mode 100644
index 0000000..9d26650
--- /dev/null
+++ b/examples/12_function_calling/function_calling.go
@@ -0,0 +1,106 @@
+package deepseek_examples
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"os"
+	"time"
+
+	"github.com/cohesion-org/deepseek-go"
+)
+
+// When you provide multiple utility functions, the model may call multiple utility functions
+// in one request and multiple times in one conversation. Specifically, the model may pass the
+// result of the previously called utility function as input to the second called utility function.
+//
+// Example:
+//   User:      What the weather and relative humidity are at the current location?
+//   Assistant: call GetLocation()
+//   Tool:      "Beijing"
+//   Assistant: call GetTemperature("Beijing") and call GetRelativeHumidity("Beijing")
+//   Tool:      "11℃", "35%"
+//   Assistant: The current temperature and humidity in Beijing are 11℃ and 35%.
+//
+// Yes, this happens in one request, so you can use it to solve complex problems.
+
+var toolGetTime = deepseek.Tool{
+	Type: "function",
+	Function: deepseek.Function{
+		Name: "GetTime",
+		Description: "" +
+			"Get the current date and time. The returned time string format is RFC3339. " +
+			"Be careful not to abuse this function unless you really need to get the real world time.",
+	},
+}
+
+func onGetTime() string {
+	s := time.Now().Format(time.RFC3339)
+	return "current time: " + s
+}
+
+func FunctionCalling() {
+	client := deepseek.NewClient(os.Getenv("DEEPSEEK_API_KEY"))
+
+	request := &deepseek.ChatCompletionRequest{
+		Model: deepseek.DeepSeekChat,
+		Messages: []deepseek.ChatCompletionMessage{
+			{Role: deepseek.ChatMessageRoleUser, Content: "What time is it now?"},
+		},
+		Tools: []deepseek.Tool{toolGetTime},
+	}
+	ctx := context.Background()
+	response, err := client.CreateChatCompletion(ctx, request)
+	if err != nil {
+		log.Fatalf("error: %v", err)
+	}
+
+	// will be empty
+	fmt.Println("response:", response.Choices[0].Message.Content)
+
+	// one tool call request
+	fmt.Println("tool calls:", response.Choices[0].Message.ToolCalls)
+
+	msg := response.Choices[0].Message
+	toolCalls := msg.ToolCalls
+
+	question := deepseek.ChatCompletionMessage{
+		Role:      deepseek.ChatMessageRoleAssistant,
+		Content:   msg.Content,
+		ToolCalls: toolCalls,
+	}
+	answer := deepseek.ChatCompletionMessage{
+		Role:       deepseek.ChatMessageRoleTool,
+		Content:    onGetTime(),
+		ToolCallID: toolCalls[0].ID,
+	}
+
+	messages := request.Messages
+	messages = append(messages, question, answer)
+	toolReq := &deepseek.ChatCompletionRequest{
+		Model:    request.Model,
+		Messages: messages,
+
+		// It is not recommended to use it unless it is a special case.
+		// The official said that they are actively fixing the problem
+		// of infinite loop calls.
+
+		// Not using this field will force the model to call the
+		// utility function only once per conversation.
+
+		// Don't try to delete only the utility functions which may cause
+		// infinite loop calls. I have tried this and the model will still
+		// call the deleted utility functions unless you delete all the
+		// utility functions, like the commented out code.
+
+		// Tools: request.Tools, // This is the key to implement chain calls.
+	}
+
+	response, err = client.CreateChatCompletion(ctx, toolReq)
+	if err != nil {
+		log.Fatalf("error: %v", err)
+	}
+
+	// will return the current time
+	fmt.Println("response:", response.Choices[0].Message.Content)
+}