Perplexity returned 429 Too Many Requests?

Perplexity documents tier-based rate limits and continuous token refill. A 429 is a scheduling signal: stop the burst, preserve any local agent work, and retry with bounded exponential backoff and jitter.

Preserve first: Record the Perplexity API endpoint, account tier, status, and response details. Pause rapid retries; if retrying is appropriate, use bounded exponential backoff with jitter and inspect local agent changes first.

Match the failure before retrying.

The request burst exceeded the current tier

Check the API console for the current usage tier and endpoint-specific rate limit.

The client is retrying without jitter

Synchronized retries can keep colliding with the same limit. Perplexity recommends exponential backoff with jitter.

An agent changed files before the API failure

Inspect the repository before resubmitting the whole task, because local tool calls can outlive the model response.

Safe recovery path.

Record the endpoint, model, request rate, status code, and any Retry-After or SDK error details.

Capture git status, git diff, changed files, and the last verified check.

Reduce concurrency and retry with capped exponential backoff plus jitter.

Resume from the preserved packet rather than asking the next request to rediscover the repository.

What the next AI should receive.

Perplexity endpoint and account tier, response status/body/request ID, current usage if shown, retry sequence and delay, local files changed by the agent, current diff, and the next allowed request.

How ShardStitch works with Perplexity.

ShardStitch supplies a Perplexity-ready clipboard packet or a packet for another agent. It preserves the project work independently of the rate-limited request and separates verified state from inferred continuation.

Source trail.

The linked Perplexity Rate Limits and Usage Tiers page describes endpoint/tier-specific limits; the SDK error-handling page discusses retries. Verify the current endpoint and tier, and use bounded backoff rather than assuming one fixed request budget.

FAQ.

Can I use one Perplexity request limit for every endpoint?

No. Check the current endpoint and account tier. The linked documentation describes tier-based limits, which can differ by endpoint.

Why add jitter to exponential backoff?

Jitter avoids clients retrying in synchronized bursts. Keep retries bounded and stop when the response or current documentation indicates that retrying is not appropriate.

Should I resubmit the entire agent task after a 429?

Not before checking files. An agent may have completed local tool actions before the API response failed. Inspect the diff and continue only with the unfinished work.

Scope note: This guide is for Perplexity API 429 responses. It does not describe rate limits in the Perplexity consumer product or guarantee a particular wait/reset time.