Recently, the majority of my prompts to AI-FDE on dev tier are getting rate limiting errors. Any tips for reducing AI-FDE rate limiting errors? I’ve tried multiple Claude Opus and Sonnet models as well as the Gemini Flash models.
The rate limiting might be due to the size of the prompt/history too. If it is too large, you might hit the limit in one hit.
Make sure you are compacting the history and remaining to something small.
You can use tricks for the output tokens (caveman, etc.) but inputs tokens are really driven by the history and prompts to the model.
You can as well .. just tell it that it has “limited context/rate” and so it will avoid expensive operations (was really day and night in some cases).
Just to add - having lots of tools or skills enabled can also increase the size of the prompt, so you could try disabling some of these
Adding to that - can avoid passing/referring big datasets/ entir log files in chat prompt. General suggestion is to pass subsets/ schema definitions in input payloads that keeps playloads small.