RemoteRunnable is not compatible with streamEvents when it's a RunnableSequence.

[errajibadr]

Issue with Streaming Responses from LangServe Endpoint in JavaScript

Hello LangChain.js community,

( i 'm not a js dev but need this for my frontend chatbot :) )

I'm experiencing an issue when trying to consume streaming responses from a LangServe endpoint using JavaScript. I have a working Python implementation, but I'm struggling to replicate the same functionality in JavaScript.

Context

• I have a LangServe endpoint running on localhost:8000.
• The endpoint is set up with a graph that includes tool calls, which are handled locally in the graph.py file.
• My Python test script works correctly, receiving content in the stream regardless of whether tool calls are used or not.

Current JavaScript Implementation

Here's a simplified version of my current JavaScript code:

import { RemoteRunnable } from '@langchain/core/runnables/remote';

const chatbot = new RemoteRunnable({url: "http://localhost:8000/"});

async function* promptAi(messages) {
  const config = { configurable: { thread_id: Date.now().toString() } };

  try {
    const stream = await chatbot.streamEvents(
      { messages: messages },
      { ...config, version: "v1" }
    );

    for await (const event of stream) {
      if (event.event === "on_chat_model_stream" && event.data?.chunk?.content) {
        yield event.data.chunk.content;
      }
    }
  } catch (error) {
    console.error("Error in promptAi:", error);
    yield "An error occurred while processing your request.";
  }
}

async function chatWithBot(userInput) {
  const messages = [{ type: "human", content: userInput }];
  let fullResponse = "";
  for await (const chunk of promptAi(messages)) {
    fullResponse += chunk;
  }
  return fullResponse;
}

The Issue

When I make a request that involves a tool call, I don't receive any content in the JavaScript implementation, whereas the Python version works fine.

For simple queries without tool calls, both implementations work as expected.

when i try an invoke i get all the graph node call in an object. the last object being the response.

Questions

1. Is there a difference in how JavaScript and Python handle the streaming responses from LangServe, particularly when tool calls are involved?
2. Are there any known issues or gotchas when using RemoteRunnable with LangServe endpoints that include tool calls?
3. Is there a recommended way to handle tool calls in the JavaScript client when they're processed locally on the server?

Any insights or suggestions would be greatly appreciated. Thank you in advance for your help!

[errajibadr]

up ?

[errajibadr]

any update on the matter?