Records & rate limits

Your plan is one allowance, counted in records, with two rolling windows — the last 5 hours and the last 7 days — the way AI chat plans work. Every company or person we show you is one record, every time it is shown: a search returning 50 companies takes 50, the people at them one each, an export row one. Counting, filters, page text and e-mail find / verify are free.

What each activity costs

A page that would show more records than your allowance has left shows what fits and says so (warnings, records_withheld) — it never overdraws.

These are the current default weights. Your organization’s weights are always in get_usage → meter, and they may be fractional (a weight of 0.1 means ten calls cost one record). A call that fails costs nothing, and a retried call or job is never charged twice.

The two windows

Both windows roll: each charge ages out 5 hours or 7 days after it was made, so the allowance trickles back continuously instead of resetting at midnight. A call must fit in both windows.

• The 5-hour window is a speed bump — it stops an agent stuck in a loop or a scraper, not a person doing their job.
• The weekly window is the plan. Unlimited has no 5-hour window, only the weekly one.

These are the defaults; your organization’s current values are always in get_usage, including any individual terms (listed as override layers). An export is capped at what is left in the tighter window, so it never overdraws.

At a limit

A call that does not fit gets quota_exceeded with the window, the numbers and the moment enough frees up — exact, computed from your own usage:

{
  "error": {
    "code": "quota_exceeded",
    "message": "5-hour allowance used up (250/250 records) — enough frees up at 2026-09-23T17:42:10Z (in 38 min); the window is rolling, so it trickles back continuously",
    "retry_after": 2280,
    "details": { "window": "5h", "limit": 250, "used": 250, "left": 0, "needed": 1, "resets_at": "2026-09-23T17:42:10Z" },
    "request_id": "req_…"
  }
}

Over the per-minute rate or the concurrency limit you get rate_limited with retry_after in seconds. Over REST both come with a Retry-After header.