Skip to main content
Version: v1.x

LangServe Endpoints And The Data Transfer Mode

LangServe Endpoints

When a runnable chain is served via LangServe, several endpoints are available for interacting with it. As of version v0.0.41, a total of 8 endpoints are available for each runnable chain: 5 endpoints for invoking and streaming the runnable, and 3 endpoints for getting the schema of the runnable.

Those endpoints are not always exposed. Developers can opt to expose only the endpoints they need for their use case. For more information on how to LangServe works and how to expose endpoints, please refer to the LangServe Documentation.

Endpoints Consumed by NLUX

In order to build an AI chatbot with NLUX, your LangServe API should expose one (or both) of the following endpoints:

  • POST /my_runnable/invoke - Invoke the runnable on a single input and return the output in a single response.
  • POST /my_runnable/stream - Invoke the runnable on a single input and stream the output as it is generated.

The following endpoint can be used by NLUX when it's available, but it's not mandatory:

  • GET /my_runnable/input_schema - JSON schema for input to the runnable.

Runnable URL Configuration

The url property is the only mandatory property for the LangServe adapter. It should include the URL to your LangServe runnable API without any endpoint at the end.

Example:

  • Good ✅ ― https://pynlux.api.nlux.ai/pirate-speak
  • Bad ❌ ― https://pynlux.api.nlux.ai/pirate-speak/invoke
import {useChatAdapter} from '@nlux/langchain-react';

const adapter = useChatAdapter({
url: 'https://pynlux.api.nlux.ai/pirate-speak'
});

Data Transfer Mode

The default data transfer mode for NLUX LangServe adapter is stream. This means that the output is streamed as it is generated by the runnable. This allows for faster response times and better user experience, as the user can see the output as it is generated.

If your endpoint only exposes the POST /my_runnable/invoke endpoint, you can still use the NLUX LangServe adapter, but you will need to set the data transfer mode to fetch.

The table below summarizes the data transfer, the endpoints that should be exposed, and the expected behavior:

Data Transfer ModeLangServe EndpointExpected Behavior
stream (default)        POST /my_runnable/streamOutput is streamed as it is generated by the runnable.
fetchPOST /my_runnable/invokeOutput is returned in a single response once the runnable has finished processing the input.
Important

In order for the NLUX LangServe adapter to work in stream mode, the LangServe API should expose the POST /my_runnable/stream endpoint. If your LangServe API only exposes the POST /my_runnable/invoke endpoint, you should set the data transfer mode to fetch.

Example: LangServe Adapter With fetch Data Transfer Mode

The example below shows how to use the NLUX LangServe adapter to connect to a LangServe API.

  • LangServe Runnable URL used: https://pynlux.api.nlux.ai/pirate-speak
  • Data transfer mode used: fetch ― The reply is returned in a single response once the runnable has finished processing the input (longer response time).
import {AiChat} from '@nlux/react';
import {useChatAdapter} from '@nlux/langchain-react';
import '@nlux/themes/nova.css';

export default () => {
  // LangServe adapter that connects to a demo LangChain Runnable API
  const adapter = useChatAdapter({
    url: 'https://pynlux.api.nlux.ai/pirate-speak',
    dataTransferMode: 'fetch'
  });

  return (
    <AiChat
      adapter={adapter}
      personaOptions={{
        bot: {
          name: 'FeatherBot',
          picture: 'https://nlux.ai/images/demos/persona-feather-bot.png',
          tagline: 'Yer AI First Mate!',
        },
        user: {
          name: 'Alex',
          picture: 'https://nlux.ai/images/demos/persona-user.jpeg'
        }
      }}
      layoutOptions={{
        height: 320,
        maxWidth: 600
      }}
    />
  );
};

Try It Out

You can edit the live code of the example above to switch the dataTransferMode to stream and notice the difference.