Design an agent's retry/backoff logic and account-tier expectations around Perplexity's usage-tier-based rate limits.

Documented steps

1. Understand that usage tiers (0-5) are driven by cumulative API spend and gate both rate limits and beta feature access: Tier 0 at $0, Tier 1 at $50+, Tier 2 at $250+, Tier 3 at $500+, Tier 4 at $1,000+, Tier 5 at $5,000+ (https://docs.perplexity.ai/docs/admin/rate-limits-usage-tiers).
2. Tier progression is automatic and permanent — once you cross a spend threshold you keep that tier's limits going forward, with no downgrade.
3. Sonar API limits are requests-per-minute and vary by model — e.g. sonar-pro is 50 RPM at Tier 0, scaling up to 4,000 RPM at the top tier.
4. Agent API limits are queries-per-second — 1 QPS (Tier 0) up to 33 QPS (Tiers 4-5); Search API is flat at 50 requests/sec with 50-request burst capacity across every tier; Embeddings API ranges 85-335 QPS by tier.
5. On exceeding a limit, expect an HTTP 429 (Too Many Requests); the docs describe a leaky-bucket algorithm where tokens refill continuously rather than resetting on a fixed window.
6. Build retry logic around 429s with backoff, and track your own request rate per-model against the documented Sonar RPM table rather than assuming one global limit covers all Perplexity APIs.