Chat Memory Stops Working When Using Custom Model Naming Configuration

[labidiaymen]

Issue Summary

Chat memory (@MemoryId) stops working when using custom named model configuration (e.g., quarkus.langchain4j.openai.default.*). The memory works correctly with the standard unnamed configuration but breaks when switching to named model configuration patterns.

Environment

• Quarkus Version: 3.28.4
• Quarkus LangChain4j BOM Version: 3.28.4
• Java Version: 24
• LangChain4j Extensions:
  • quarkus-langchain4j-core
  • quarkus-langchain4j-openai
  • quarkus-langchain4j-pgvector
  • quarkus-langchain4j-mcp
  • quarkus-langchain4j-agentic
• Build Tool: Maven
• OS: macOS

AI Service Interface

@RegisterAiService(
    retrievalAugmentor = RagRetriever.class,
    modelName = "default"
)
public interface AssistantService {

    @McpToolBox
    @ToolBox(TimeTools.class) 
    @SystemMessage("{systemMessage}")
    @UserMessage("{userMessage}")
    Multi<String> generateChatResponse(
        @MemoryId String conversationId, 
        String systemMessage, 
        String userMessage
    );
}

Working Configuration (Standard Unnamed Model)

When using the standard unnamed configuration, chat memory works correctly:

# Standard unnamed model configuration
quarkus.langchain4j.chat-model.provider=openai
quarkus.langchain4j.openai.api-key=sk-proj-***
quarkus.langchain4j.openai.chat-model.model-name=gpt-4

# Memory configuration - WORKS with standard unnamed model
quarkus.langchain4j.chat-memory.type=MESSAGE_WINDOW
quarkus.langchain4j.chat-memory.memory-window.max-messages=20

Broken Configuration (Custom Named Model)

When switching to custom named model configuration (e.g., using .default. prefix), chat memory stops working:

# Custom named model configuration using .default.
quarkus.langchain4j.chat-model.provider=openai
quarkus.langchain4j.default.chat-model.provider=openai
quarkus.langchain4j.openai.default.api-key=sk-proj-***
quarkus.langchain4j.openai.default.chat-model.model-name=gpt-4

# Attempted memory configuration 1 - DOESN'T WORK
quarkus.langchain4j.chat-memory.type=MESSAGE_WINDOW
quarkus.langchain4j.chat-memory.memory-window.max-messages=20

Alternative Memory Configuration Attempts (All Failed)

# Attempt 2 - Using .default. prefix for memory - DOESN'T WORK
quarkus.langchain4j.default.chat-memory.type=MESSAGE_WINDOW
quarkus.langchain4j.default.chat-memory.memory-window.max-messages=20

Is there any documentation or example showing:

• Custom named model configuration + chat memory working together
• Multiple named models with separate memory configurations

Thanks

[geoand]

Thanks for reporting!

Any chance you can put together a sample that we can use to test potential fixes against?

[geoand]

Actually, I took a closer look, and it seems like we don't support a "named" configuration for the chat memory. I am not even sure what that name would correspond to TBH.

@jmartisk WDYT about this?

[jmartisk]

I guess it could have some use - if some AI services need to use a message window memory, some may use a token window memory or a different window size...
The @RegisterAiService annotation would have to receive a new parameter, chatMemoryConfigurationName or so, that allows specifying which chat memory config should be used, similar to modelName that it already has.
It's hard to guess how complex and intrusive it would be to implement at this point though.

[bitsofinfo]

experiencing this as well, chatmemory will just recreate constantly for the same memoryId when using

@RegisterAiService(chatMemoryProviderSupplier=

and a custom named ai service....

[geoand]

@bitsofinfo can you be a little more specific please as I don't really understand your comment?

Thanks