>unstuck engine

> Rate limits

Limits apply per workspace to programmatic traffic — API keys and connected assistants together, however many keys or connectors you have. Using the app itself is never counted. Limits are part of the plan:

Plan Requests per minute Requests per hour*
Free 30 500
Pre-PMF 1,500 / 4,000 120 3,000
Pre-PMF 10,000 and up 300 10,000
Agency / custom unlimited unlimited

In a project, every workspace is limited by the paying workspace's plan — counted per workspace, or in one bucket shared by the whole project, as set for your project. A workspace without an active plan is limited like Free.

* The hourly window is a second layer that is currently off; the minute window is always on. Values are data, not code — they can change per plan without a release, and this table is the reference at the time of writing.

How you see it

Once a limit applies, every response carries:

X-RateLimit-Limit: 30          # the tightest window's limit
X-RateLimit-Remaining: 27
X-RateLimit-Window: minute     # minute | hour
X-RateLimit-Reset: 60          # seconds until the window has surely moved on

When the window is spent the call returns 429 rate_limit_exceeded with Retry-After: <seconds> and the same headers; the message says how long to wait. Nothing was executed. The CLI prints a note on stderr when fewer than 20% of the window's requests remain; MCP tools append the same sentence to the tool result so the assistant slows down on its own.

What counts

One customer-initiated request = one unit. An MCP tool call is one request even though it fans out to an endpoint internally; initialize and tools/list from your assistant count too (they are cheap and rare). Requests refused before authentication (no or invalid key) are not counted against the limit.

Staying under

  • Page bigger, call less: page_size up to 500 on list-leads / list-companies.
  • Filter server-side (stage, since/until, ids) instead of scanning.
  • Batch in one process rather than many parallel scripts.
  • Need more? Plan upgrade, or talk to us about custom terms.