# 429: you are being rate limited

## The error

`429 Too Many Requests: You are being rate-limited.`

## Fix

1. **Back off with jitter.** Exponential backoff (base delay doubling, capped, plus random jitter) is the documented pattern. The error-handling guide includes a full implementation; the key principles are jitter, max retries, capped delay, and never retrying other 4xx errors.
2. **Batch.** Fewer, larger requests beat many small ones. Upsert in batches up to the 2 MB request limit instead of one vector per call.
3. **Monitor.** Watch request metrics and reduce rate proactively as you approach limits. Most limits can be raised on request; contact support with the use case rather than engineering around a cap in code.

## Rules for agents

1. 429 is the one 4xx you retry. Every other 4xx is a bug in your request; retrying it is a loop.
2. Put the backoff in the client layer once, not around every call site. Scattered sleep calls drift and compound.
3. For sustained high throughput, size it properly: dedicated read nodes remove read-unit limits for query, fetch, list, and full-text search. See the throughput guides before building a client-side throttle maze.