I've seen people using databases or

[opal basin]

I had good experience just now by making my agent remember everything, then I turn on autocompaction (in docs) but Idk if it worked automatically so then I told him, to write down everything important and my MEMORY.md just went wild, about 117k of novel writing context was remembered with references to the source documents, he did it really well. Then I ran /compact <msg> with msg being "preserve everything about the novel we are writing" and that compacted down to about 23k so my calls are back to costing 1-2 cents again (instead of a quarter each)

[hardy saddle]

Thats pretty cool thanks for sharing

So u gave it access to a gdocs and had it maintain things there?

[opal basin]

havent tried that specific case no, but I flexed the builtin system pretty good and I'm hoping autocompaction keeps me below 80k context

[hardy saddle]

How did it affect the memory and performance when context window gets full