langchain-ai/langchainjs

AI Summary: The user is experiencing an issue where `response_metadata` is always empty when using `streamEvents` with `ChatVertexAI` and `createReactAgent`. While `.stream()` provides the desired metadata, it makes streaming to the frontend difficult. The user needs a way to stream the response to the client and simultaneously capture final metadata, such as errors or finish reasons, once the stream completes.

AI Summary: The task is to debug a LangChain.js integration with AWS Bedrock Claude. The issue is that custom metadata isn't being passed to the Bedrock invocation, despite being supported by both LangChain and the Bedrock SDK. The problem lies within LangChain's internal handling of metadata, specifically the deletion of the `metadata` property before the Bedrock call. A solution needs to be implemented that conditionally adds the metadata to the request, only when the underlying model supports it (like Bedrock Claude), without breaking compatibility with other models.

AI Summary: Extend the LangChain.js Groq integration (@langchain/groq) to support multimodal inputs, specifically images, enabling the processing of messages containing both text and image content. This requires modifying the `convertMessagesToGroqParams` function to handle non-string message content, likely by adding a check for image data and adapting the Groq query accordingly.