Fix: Incorrect chat title generation

[sh4shii]

Description

fixes: #4774
The chat title was previously being generated as [object Object][object Object]. This issue has been fixed by ensuring the correct string representation is used for the title.

Checklist

• [] The relevant docs, if any, have been updated or created
• [] The relevant tests, if any, have been updated or created

Screenshots

As you can see in the attached screenshot, the chat title is now generated correctly.

Testing instructions

[ For new or modified features, provide step-by-step testing instructions to validate the intended behavior of the change, including any relevant tests to run. ]

[netlify]

✅ Deploy Preview for continuedev canceled.

Name	Link
🔨 Latest commit	ad0667e
🔍 Latest deploy log	https://app.netlify.com/sites/continuedev/deploys/67e0678cc133250008d86deb

[RomneyDa]

@sh4shii would it be possible to implement the same fix using chat instead of complete? Something in the latter lines you updated is likely the culprit?

Specifically wondering what the [object Object] actually is in your case. Is it an error because the model doesn't support chat? Is it not parsed correctly?

[sh4shii]

@RomneyDa tried implementing the fix using model.chat, but it's still returning [object Object][object Object] instead of the expected string.

It seems like the response from model.chat is not being parsed correctly, as it returns:
titleResponse
{role: 'assistant', content: '[object Object][object Object]'}
Whereas model.complete correctly returns a string with the title.

This suggests that either the model's chat response structure differs, or additional parsing is needed when using model.chat. Do you have any insights on why this might be happening? I can explore further based on that.

[RomneyDa]

I don't have an immediate answer to why chat is returning {role: 'assistant', content: '[object Object][object Object]'} but let's debug that before switching to complete. If you put a breakpoint in chat for whichever provider class you're using (or BaseLLM) you should be able to walk through and find this. I haven't been able to duplicate on my end, would you be able to try that breakpoint?

[sh4shii]

@RomneyDa I've investigated with breakpoints in the Gemini model class as suggested. Here's what I found:

1. Within the stream chat, the message structure looks correct:
   message:
   {role: 'assistant', content: Array(1)}
   content: (1) [{…}]
   0: {type: 'text', text: 'React'}

2. However, when this response is returned as titleResponse, it somehow transforms to:
   {
   role: 'assistant',
   content: '[object Object][object Object]'
   }

Would you be able to point me to which specific files/methods I should add logging or breakpoints to track this transformation?

[RomneyDa]

@sh4shii looks like could be one of two lines. Probably BaseLLM's chat function:

async chat(
    messages: ChatMessage[],
    signal: AbortSignal,
    options: LLMFullCompletionOptions = {},
  ) {
    let completion = "";
    for await (const chunk of this.streamChat(messages, signal, options)) {
      completion += chunk.content;
    }
    return { role: "assistant" as const, content: completion };
  }

In completion += chunk.content, chunk.content is MessageContent, which can be an array of parts. I might try that first

Same issue actually shows up in ChatDescriber

removeQuotesAndEscapes(titleResponse.content.toString());

where content can be an array

[RomneyDa]

@sh4shii done here #5425
Added you as coauthor on the commit, thanks for the contribution!