[google-adk-spring-ai] MessageConverter.toLlmResponse does not map ChatResponse Usage to LlmResponse

[caliu1]

contrib/spring-ai/src/main/java/com/google/adk/models/springai/MessageConverter.java

🔴 Required Information

Describe the Bug:
com.google.adk.models.springai.MessageConverter.toLlmResponse(ChatResponse, boolean) drops Spring AI ChatResponse usage metadata.
Even when chatResponse.getMetadata().getUsage() is present and non-zero, the returned LlmResponse has empty usageMetadata.
This breaks token accounting in downstream ADK plugins/callbacks.

Steps to Reproduce:

1. Use google-adk-spring-ai in a Java project with streaming chat.
2. Enable streaming usage on Spring AI model options (e.g. OpenAiChatOptions.builder().streamUsage(true)).
3. Get a ChatResponse where chatResponse.getMetadata().getUsage() is non-null and has non-zero values.
4. Convert it via MessageConverter.toLlmResponse(chatResponse, true).
5. Check llmResponse.usageMetadata() -> it is empty / missing.

Expected Behavior:
LlmResponse.usageMetadata should be populated from ChatResponse.getMetadata().getUsage()
(e.g. prompt/completion/total token counts), so ADK callbacks/plugins can read usage reliably.

Observed Behavior:
LlmResponse contains content/partial/turnComplete, but no usageMetadata.
As a result, token metrics remain 0 in usage-dependent logic.

Environment Details:

• ADK Library Version (see maven dependency): google-adk-spring-ai:0.9.0 (also google-adk:0.9.0)
• OS: Windows
• TS Version (tsc --version): N/A (Java project)

Model Information:

• Which model is being used: qwen-plus (OpenAI-compatible endpoint via DashScope)

---

🟡 Optional Information

Logs:

ChatResponse metadata usage is present and non-zero before conversion.
After MessageConverter.toLlmResponse(...), LlmResponse.usageMetadata is empty.
Downstream token counters therefore record 0.

Additional Context:
In the current implementation, toLlmResponse(ChatResponse, boolean) builds LlmResponse with:content、partial
and turnComplete，but does not map usage from Spring AI response metadata.

Minimal Reproduction Code:
Please provide a code snippet or a link to a Gist/repo that isolates the issue.

package repro;

import com.fasterxml.jackson.databind.ObjectMapper;
import com.google.adk.models.LlmResponse;
import com.google.adk.models.springai.MessageConverter;
import org.junit.jupiter.api.Test;
import org.springframework.ai.chat.messages.AssistantMessage;
import org.springframework.ai.chat.metadata.ChatResponseMetadata;
import org.springframework.ai.chat.metadata.DefaultUsage;
import org.springframework.ai.chat.model.ChatResponse;
import org.springframework.ai.chat.model.Generation;

import java.util.List;

import static org.junit.jupiter.api.Assertions.*;

class MessageConverterUsageReproTest {

    @Test
    void usageShouldBeMappedFromSpringAiChatResponseToAdkLlmResponse() {
        MessageConverter converter = new MessageConverter(new ObjectMapper());

        AssistantMessage assistantMessage = new AssistantMessage("hello");
        Generation generation = new Generation(assistantMessage);

        DefaultUsage usage = new DefaultUsage(12, 8, 20);
        ChatResponseMetadata metadata = ChatResponseMetadata.builder()
                .id("resp-1")
                .model("dummy-model")
                .usage(usage)
                .build();

        ChatResponse chatResponse = new ChatResponse(List.of(generation), metadata);

        // Precondition: Spring AI usage exists and is non-zero
        assertNotNull(chatResponse.getMetadata());
        assertNotNull(chatResponse.getMetadata().getUsage());
        assertEquals(20, chatResponse.getMetadata().getUsage().getTotalTokens());

        LlmResponse llmResponse = converter.toLlmResponse(chatResponse, true);

        // Expected: should be present, but currently missing in google-adk-spring-ai
        assertTrue(llmResponse.usageMetadata().isPresent(),
                "Expected usageMetadata to be mapped from ChatResponse metadata usage");
    }
}

How often has this issue occurred?:

• Always (100%)

[hemasekhar-p]

Hi @caliu1, Thank you for reporting this issue. I have created the PR #1161 to resolve the missing token tracking in the adk-java-springai module. The fix updates MessageConverter to correctly map Spring AI's usage metadata into the ADK's LlmResponse. hope this will resolve your problem?

[abhadra2]

Hi - I am also facing the same issue , can I have a build to test this out ?