Skip to main content
Exceeding a limit returns 429 with error.type: rate_limited and a Retry-After header. Retry-After is the number of whole seconds until the counter that refused you resets (at least 1).

Per API key and per workspace

Every /v1 call, and every MCP tool call, counts against the API key that made it and against its workspace. These are the defaults: Each key has its own counters, so one busy key does not use up another key’s allowance. A request refused by its key’s limit does not count against the workspace. Different workspaces never share these counters. Contact support if your workspace needs different limits.

Per destination prefix

POST /v1/otp/send is also counted per destination prefix: the first six digits of the E.164 number (country code plus the start of the network code). The default is 300 OTPs per prefix per hour for each workspace. A sudden burst of codes to one number range is a common sign of SMS pumping.

SMS pumping protection

SMS pumping is traffic that requests codes for numbers the attacker is paid for, not for real users. Robase has these controls against it:
  • Countries. Each workspace has a list of countries it sends to, under Settings → Countries. New workspaces start with their billing country. Workspaces that existed before this control started with every country they had sent to in the previous 90 days. For now, a send to a country that is off is logged but still sent. When enforcement starts, it will be refused with 403 country_not_enabled.
  • Unusual prefix traffic. Every five minutes, Robase checks each destination prefix. A prefix is flagged when, in the last hour, it had at least 30 OTPs, under 20% of them were verified, and the volume was at least three times the prefix’s usual hourly volume over the last 7 days. The first time, the prefix is throttled for an hour: at most 10 OTPs an hour to that prefix. If it is flagged again within a day, it is blocked for your workspace. Sends to it return 403 prefix_blocked until support lifts the block. These checks can miss some attacks and can flag real traffic. Contact support if a prefix you need is blocked.

Per destination phone

These counters are kept per workspace and per E.164 number. Your traffic to a number never counts against another workspace sending to the same number. POST /v1/sms/send uses the same intake check as OTP: a Benin number that is not 10 digits starting with 01 is refused before this counter, and nothing is charged. The POST /v1/otp/send limit is a fixed window: it starts with the first send to that number and resets 10 minutes later, however many requests arrive in between. A request counts once its body, its options and the workspace’s standing have been checked and the number parses as E.164 and passes the numbering-plan check. From that point it counts even if it is then refused, for example with 402 insufficient_credits, 400 country_not_supported, 403 prefix_blocked, or a 429 from this limit itself. Requests refused before that point (an invalid body or number, including a Benin number that is not 10 digits starting with 01, a channel your delivery mode does not allow, or a workspace that is held for verification, suspended or restricted) do not count. A new POST /v1/otp/send to the same number does not cancel an earlier code. Each send returns its own id and code, and an earlier code stays valid against its own otp_id until it expires, is verified, or uses up its 5 attempts. POST /v1/otp/verify checks the code against the otp_id you send, so store the newest id on every resend and verify against that one.

Per client IP

A general bucket of 1200 requests per minute per client IP covers /v1 and the dashboard (internal/router.go). Auth endpoints and provider DLR callbacks have their own limiters and are exempt from this bucket. Successful responses may include X-RateLimit-Remaining. On /v1 it is the smaller of what is left for the key and for the workspace this minute.

Back off

Honour Retry-After. Do not tight-loop a 429. OTP send is intentionally tight so a leaked key cannot flood a single number. Over MCP, a rate-limited tool call returns error.type: rate_limited with error.retry_after_seconds.