That's what was in my OP, I used the API approach, but the rate limits were insanely low (seemingly?) the agent died after 3 steps in a single question.
gpt-5-2025-08-07
38.887K input tokens
That was my usage, and I got rate limited. Thank you for your tips!