From waht I understand, the transcript has the lines, the summary a VERY short summary (which is really short, given that GPT 4.5 now has a16k model). Anyhow, those overlap. One mode that I have been exploring in code, is combining this approach. * There is a sumamry generated, but.... * When building, I go like this: Work backward in the history. Take the transcript for X turns, then go on until you find a summary (which I do only generate every 4th turn), adding transcript lines and finish with the transcript. Ok, not really - my appraoch ahs 3 stages: Transcript, Journal (where the AI does a simple user said / i said) and summary. Anhow. I asked in help and I was told that the sumamry is always current - which means that summary and transcript overlap (not helping the AI and wasting space). A more detailed system may work nicely. It also may not need to be used via variable - I have good success inputting it in the context outside of the direct message, constructing a history.