the codex tokens get depleted very fast on simple tasks like edit a file and it costs so much on my

[oblique ridge]

so i use codex for mtasks like coding but am disappointed on how the tokens get depleted very fast and costs me dearly

[dreamy mesa]

they do it on purpose

also dont be so heavily reliant on AI to edit a file 💔

[oblique ridge]

ok

[lusty knot]

try npx @ccusage/codex@latest to get an overview of used tokens. This might be off a little bit if you use e.g. chronicle, but it should give you a good overview.

A simple file edit shouldn't use many tokens, but it might get more if it needs more context. One factore might be a big file size.

(yes, I was typing longer, because I verify that the things I suggest definitely work)

[golden crag]

my bad for the unnecessary commentary lmfao

[lusty knot]

No worries, I often look at a screen when someone is typing and think exactly the same 😉

[golden crag]

additionally, conversation/chat length absolutely ruins chatgpt’s response time, memory and performance, and your memory usage on your computer spikes like crazy when opening a long chat

to add to your prior message

[lusty knot]

I always forget that there are still people out there keeping long chats. Just start a new one whenever anything is done. One patch/feature = one chat. And only use it again after hours/days, if you really really need that context – and know the consequences.

[golden crag]

might be a bit excessive but not a bad habit, certainly better than opening a chat and your window using 12gb of memory while it’s loading everything, happens with long chats

[remote tide]

some projects require longer chats because their nature is such. i am developing a visual grammar and using chatGPT, claude and gemini - which help refine it. so during our conversations we make certain decisions with a version of some aspect of the grammar, which later might change. but i cannot simply make a new chat when that happens because there is a lot of context behind each decision, that then needs to be copied over or re-iterated to make it a live memory for the AI agent to remember and that is a waste of tokens for re-establishing the truth. but i have to do that from time to time once the thread becomes long enough to have responsiveness issues... i have a solution for it, but i haven't a clue how to get someone from OpenAI to get eyes on it. 😉

[lusty knot]

This sounds like something which should be documented in the code base, but maybe I don’t understand it enough and you found a niche there. Take my comment as "in general" or "for most users".