Rate limiting and abuse prevention

Fair use, enforced

Rate limits protect availability and cost. Limiting per authenticated identity is far more accurate than per ip address, since many legitimate users share addresses and one attacker can use many.

Communicate the limit. Returning the remaining quota and a retry-after header lets well-behaved clients back off correctly instead of hammering you into a longer ban.

Expensive endpoints deserve their own limits. Search, export, and anything invoking a model cost far more per call than a simple read, and a global limit prices them identically.