[Bug]: The promptPrefix (system message) seems to get truncated when conversation exceeds the context limit.

[Lavanille777]

What happened?

As the conversation approaches the context limit, I noticed that the large model seems to forget its system prompt. I haven’t confirmed this through packet capture yet, but it appears to be the case. The version I am using is 'librechat-dev:8b31f255f54a570c794d61cef313d1c43a7cd356'

Steps to Reproduce

With gpt-4o-2024-11-20, you can command it to output the first sentence of its system prompt. Then, as your conversation exceeds the maximum context length, you’ll notice that it can no longer recall or reproduce the system prompt.

What browsers are you seeing the problem on?

No response

Relevant log output

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

This was working when system instructions were appended before the latest message. I changed this to move the system message placed first in the chat history to conform to what a lot of providers expect (namely DeepSeek, Mistral), and forgot to update this. Will fix soon