After starting to use Claude Code to develop external integrations, I asked @pierre-gilles if there weren’t any small tips to optimize my quota, which is being used up very quickly.
Thanks to @pierre-gilles for his time and recommendations.
Here’s what I took away:
Create a CLAUDE.md file at the root of the project (Run /init in Claude Code on the repo)
Have one conversation per topic. Basically, one conversation per bug/feature. Otherwise, a simple « hello » in a topic with 400K of context will force the AI to reread the entire conversation and consume 400K tokens
Be careful with the PR monitoring feature, which consumes a lot (Make sure to disable it globally and verify that Claude doesn’t enable it on each conversation)
Globally, it’s here:
You can’t really think of it that way. The key variable is the time you’ll take to develop it.
In fact, when you take a subscription, you have token quotas per 5-hour slot and per week.
So you have to juggle with that. You can, in one big request, use up all your Fable 5 tokens in 1 hour… so you have to wait 4 hours to continue. But if you do that 5 or 6 times in a row, you’ll have burned through your weekly quota, so you’re stuck for a longer period.
If you develop calmly, you’ll be able to develop as many integrations as you want, but in small chunks.
You can also use the /compact command when it is really not possible to start a new conversation.
This command condenses the exchange history into a concise summary to reduce the amount of active context used. Old messages are replaced by an overview grouping important decisions, file changes, and essential code structures.
