Mito AI: Restore chat on refresh #1488
Conversation
Rename full_message_history to llm_message_history for clarity
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
clear file when clearing history
| data = { | ||
| "llm_history": self._llm_history, | ||
| "display_history": self._display_history, | ||
| } |
There was a problem hiding this comment.
Another option here is to just store the metadata that we send from the frontend to the backend in order to create the prompt in the .json file. Then, we can easily recreate both the llm_history and the display_history from them. There's no reason for the backend to care about the display_messages for example.
What do you think about refactoring this? Is there an advantage to storing everything in the .json file itself?
There was a problem hiding this comment.
The reason for the backend to care about display_messages is to be able to restore them in the frontend when the extension loads.
I chose the format of what is stored in JSON to match what is stored in the memory, so it is easy to serialize/deserialize. That being said, we don't have to strore full text representations of either and can recalculate on the fly from context objects.
As I see it, there is no compeling reason to store that metadata, becuase the only thing we do with it is to calculate LLM prompts and display messages.
| ${input}`}; | ||
| } | ||
|
|
||
| const removeInnerThoughtsFromMessage = (messageContent: string) => { |
There was a problem hiding this comment.
I think we're going to need this sort of post-processing code in the frontend for when we stream responses to the frontend.
Description
Testing
Documentation
N/A