Rate limits vary by endpoint. The defaults below apply to most accounts. Send an email to gtm@crustdata.co to discuss higher limits if needed for your use case.
Core behavior
Crustdata applies per-endpoint request limits. Operationally, you should assume that steady request distribution is safer than bursts, even when total requests per minute look acceptable.Default rate limits by endpoint
Contact and identify endpoints
/person/contact/enrich,
/batch/person/contact/enrich, /person/identify, and
/batch/person/identify are grouped with the live endpoints, not with
in-database, so their limits do not move with the limits on search and
enrich. Raising a live limit is constrained. Contact sales or write to
gtm@crustdata.co with your volume./watch/company/search leaves
/watch/company, /watch/person/search, and /watch/job/search with a full
allowance. This bounds bursty create and update loops; steady use is unaffected.
Every watch response carries the usual counters, and the eleventh request in a
minute is refused:
429 - watch-management rate limit
Response headers
X-RateLimit-Reset counts down the seconds until the window rolls over, so wait
that long rather than retrying immediately.
The four /account/usage/* endpoints share one budget of 60
requests per minute. A burst of /account/usage/events calls also blocks
/account/usage/summary until the window resets.
Agent Search allows 6 starts per minute on
POST /agent/searches, and a refused start counts toward that limit. The
read, events, and cancel routes share one budget of 300 requests per minute.
Over either limit, the 429 has no Retry-After header, so wait for the
seconds in X-RateLimit-Reset. When too many Agent Search event streams are
open, a new stream gets 429 with Retry-After: 5.
Batch job concurrency
Batch endpoints carry a second limit on top of the per-minute rate: a cap on how many jobs you may have in flight at once. A job counts as active while itsstatus is pending or processing.
Active jobs are counted in two independent pools, so a queue of enrichment
jobs never blocks a live lookup and vice versa:
Each pool’s cap is derived from your rate limit on that pool’s endpoints: the
highest RPM you hold on any endpoint in the pool, divided by six, then held
between 5 and 30. On default limits that is 5 active jobs in each pool, and
raising your rate limit raises the cap with it. Your current per-endpoint limits
are visible in your dashboard.
Exceeding a pool’s cap returns
429 with error.type of rate_limit_error and
a Retry-After header:
429 - too many active jobs
429 here costs nothing. Poll GET /batch/{batch_id} until a job reaches
completed or failed, then submit the next one. A job that has been active
for more than 24 hours stops counting toward the limit, so a stuck job does not
block your account indefinitely.
Rarely, the platform as a whole is at batch capacity. That also returns 429
rate_limit_error, with a message saying so and a shorter Retry-After. It is
unrelated to your account’s limits - retry after the interval given.
Implementation guidance
To reduce429 Too Many Requests responses:
- Spread traffic across the full minute instead of burst-sending.
- Use retry logic with exponential backoff and jitter.
- Keep request queues bounded.
- Add circuit breakers around non-critical enrichment flows.
- Monitor request logs and alert on sustained
429rates.
Client-side best practices
- Centralize throttling in one shared HTTP client.
- Use endpoint-specific concurrency and QPS controls.
- Prioritize business-critical requests when backpressure starts.
- Cache stable results where your workflow allows it.
Rollout checklist
- Start with conservative throughput.
- Increase gradually while tracking latency and error rates.
- Recalibrate limits when you add new endpoints or workflows.

