Call check() directly in any runtime, or wrap middleware() for Express-compatible request flows.

Sliding window, fixed window, token bucket, and leaky bucket cover fairness, burst capacity, and smoothing needs.

Start with in-process Memory, then move to Redis or CacheHubStore when counters must be shared across instances.

Close owned Redis clients and cache-hub cleanup resources with await limiter.close().

Keep IP allowlist authorization separate from route quotas, with global and route-level configuration patterns.

Memory, Redis direct, HTTP middleware, and OSS comparison data are documented with commands and environment.