> ## Documentation Index
> Fetch the complete documentation index at: https://docs.landedfees.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Working with rate limits

> Practical patterns for staying under 60 requests per minute per key.

The absolute rules live in the [Rate limits reference](/rate-limits).
This guide is the how-to.

## Read the headers on every response

The engine ships `X-RateLimit-Limit`, `X-RateLimit-Remaining`, and
`X-RateLimit-Reset` on every 2xx. Log them or plumb them into your
metrics.

```ts theme={null}
const resp = await fetch(url, { headers });
const remaining = Number(resp.headers.get("X-RateLimit-Remaining"));
if (remaining < 10) metrics.warn("landedfees_ratelimit_close", { remaining });
```

## Batch before you loop

* 500 individual calc calls sent in a tight loop will 429 immediately
  (and burn 500 quota entries).
* One `/v1/calc/bulk` call with 500 rows uses one rate-limit slot and
  one quota entry.
* Six `/v1/rates` calls for a sourcing comparison burn six slots.
  One `/v1/compare` call fans out to all 6 origins server-side and
  uses one slot.

## Concurrency ceiling

Cap outbound concurrency at 8 workers per key. At 8 concurrent
workers averaging 500 ms per call, sustained throughput is roughly
960 rpm from the calling side, but only 60 rpm from the server side
will succeed. The excess simply 429s. Prefer fewer workers with the
built-in retry loop.

## Multi-key sharding

Rate limits are per key, not per organization. Split workloads by
minting one key per workload:

* `prod-quote-engine` for real-time UI quoting.
* `ci-nightly-audit` for the nightly re-run.
* `data-warehouse-etl` for the ETL that pulls rate snapshots.

Each key gets its own 60 rpm bucket. Do not mint keys purely to
dodge the limit; the monthly quota is per organization.

## Backoff with jitter

The SDKs implement exponential backoff with jitter automatically. If
you are rolling your own:

```
attempt 0: retry-after seconds + rand(100..500) ms
attempt 1: 2s  + rand(100..500) ms
attempt 2: 4s  + rand(100..500) ms
attempt 3: 8s  + rand(100..500) ms
cap at 60s. Give up after attempt 5.
```

## When to ask for higher limits

Ask when:

* Your workload genuinely exceeds 60 rpm per key **sustained**,
  after moving to bulk and compare endpoints.
* You have a burst pattern that a longer window cannot absorb.

Growth and Business ceilings are fixed at 60 rpm per key. Custom
ceilings ship on Enterprise contracts. Contact sales.