Rate limits

Each API key, and each app you connect by signing in, has its own allowance of requests per minute, set by the plan of the workspace it works in. A busy script never uses up the allowance of your assistant.

Every request a key or app makes counts, over REST or MCP, including ones that are refused. GET /me and cheqq_whoami tell you the current allowance.

How many keys each person can have is on API keys.

Watch the headers

REST and MCP answers carry what's left of this minute, and how many seconds until it refills:

X-RateLimit-Limit: 1000
X-RateLimit-Remaining: 997
X-RateLimit-Reset: 42

Past the limit

The request is refused with 429, and nothing is done. The answer says how many seconds to wait:

Retry-After: 18
X-RateLimit-Limit: 1000
X-RateLimit-Remaining: 0
X-RateLimit-Reset: 18

{
  "error": {
    "code": "RATE_LIMITED",
    "message": "This key made more than 1000 requests in a minute. Wait a moment and try again.",
    "details": { "limit": 1000, "retryAfterSeconds": 18 }
  }
}

Wait Retry-After seconds before trying again, and back off further if you're refused again.

Signing in

Apps that connect by signing in also have limits on how often one address may exchange or revoke tokens. Past one, the answer is 429 with the usual OAuth error body, { "error": "too_many_requests" }, and the same Retry-After and X-RateLimit-* headers. So does sending too many keys or tokens that aren't valid from one address.

Using less

• Write in batches. Add or change up to 200 records in one request with batch-create and batch-update.
• Count and summarize instead of reading everything. records/count and records/aggregate answer in one request.
• Ask for bigger pages. Records page up to 200 at a time.
• Give each job its own key. Allowances are per key, so a nightly sync and your assistant don't compete.

Other limits still apply where they do in Cheqq: runs count toward your workspace's monthly step runs, and AI steps use credits. See Billing.