Chain of Thought UI crashes when langchain steps output dicts instead of strings

[mpartel]

With [Async]LangchainCallbackHandler, when my intermediate steps output dicts instead of strings, the UI crashes.

Repro:

• Use the following code.
• Input a message.
• Click on "Took 1 step".
• Get "Something went wrong." with JS console saying "TypeError: r.trim is not a function".

import chainlit as cl
from langchain.chains import TransformChain

@cl.on_message
async def main(message: str):
    def transform(inputs):
        return {
            'output': {
                'structured': {
                    'report': 'Done'
                }
            }
        }
    chain = TransformChain(
        input_variables=['text'],
        output_variables=['output'],
        transform=transform,
    )
    result = chain.run({'text': message}, callbacks=[cl.LangchainCallbackHandler()])
    return await cl.Message(content=result['structured']['report']).send()

(Possibly related typing bug in LangChain 0.0.267, but I'm almost certain their intention is to allow intermediate steps to output whatever.)

[willydouhard]

Interesting, will investigate

[willydouhard]

Okay so this chain outputs a json object. Currently the Message component in the UI only accept strings. What would be the expected behavior here? Display the json object in a code block?

[mpartel]

Yes, that sounds good 👍