by dtagames on 6/22/2025, 1:12:29 PM
The reason is that each response in a thread is the response to not only the last prompt but all of the other prompts and responses in that thread. The LLM itself has no memory, so to answer every prompt, the entire thread is sent again.
This means that extracting a portion of the thread and starting a prompt from there won't give the same result and may not give a correct result at all if the excised section contains critical information.
So it is possible, and you can write your own RAG app which works that way, but you won't be happy with the results.
If you use Cursor, you can segment parts of the conversation into "plan" or "report" files (these are just markdown documents you ask it to make), then reference those files in a later conversation to bring in that context without retyping the thread.
by esperent on 6/22/2025, 3:36:42 AM
Librechat can do this - edit both your messages and the LLMs messages. I edit my own messages all the time but I've never found a reason to edit the LLMs messages.
If you just mean editing your own messages, I think all the LLM chat apps allow this.
by neximo64 on 6/22/2025, 2:11:04 AM
As far as I can see you can edit in ChatGPT, Gemini, Claude. The api playgrounds allow editing previous messages (even from the LLM), deleting, ..
What are you trying exactly, what LLM do you mean as they all have this functionality.
But many of the providers lock their best models behind their interface before the developers get access, or lock features behind them.
Why can’t we delete messages? It just seems like common sense to me. I spend way too much time “ruining” a thread then spending 20 minutes getting a new thread back to where I previously was. All because I can’t delete the messages that derailed everything.
Just being able to right click and “delete” or “edit” any message in the chat would be the biggest upgrade to user experience in my opinion. Then when the LLM inevitably hallucinates or misunderstands you don’t have to fight to get it back on track or abandon it altogether and start a new chat and potentially spend 20 minutes priming the chat.
It’s also trivially simple to implement. I wonder if the providers have some explicit reason for this? Is it to bias people to start new chats rather than have long ones which saves them money? I just feel like there has to be a reason this dead simple feature isn’t possible. In the API it’s literally just changing the messages array that is the input.
Does anyone know the reason they don’t have this? (Surely someone before has said: “we could let users delete or edit messages” and it was shot down.. right?)