Rate Limiting

The API supports a per-org sliding-window rate limit, but it is off by default — requests are unthrottled until a per-org per-minute limit is configured on your plan. When a limit is in effect, every response — successful or not — carries the current state of your budget so you can pace requests proactively instead of reacting to 429s.

The numbers

There is no fixed default limit. The ceiling is a per-org, per-minute number; while it’s unset (the out-of-the-box state) nothing is throttled and the rate-limit headers are omitted entirely. Once a limit is configured:

The limit applies to each API key separately: every key gets the full per-minute number for its environment, and one key using up its budget never throttles another. Need a limit raised, lowered, or set? Contact support.

OAuth tokens and MCP calls use the same per-org, per-minute number from their own buckets (ratelimit:oauth:… and ratelimit:mcp:…), so an agent’s exploratory burst cannot exhaust any API key’s budget in the same organization. An MCP tools/call is metered on the organization it names in its organization argument — a connection covering two organizations on different plans gets each one’s own limit.

Headers on every response

Once a limit is configured, two header families are written on every response — the modern IETF-draft RateLimit-* names and the legacy X-RateLimit-* names (identical values), so clients pinned to either draft keep working:

RateLimit-Limit: 600
RateLimit-Remaining: 412
RateLimit-Reset: 1747512000
X-RateLimit-Limit: 600
X-RateLimit-Remaining: 412
X-RateLimit-Reset: 1747512000

• RateLimit-Limit / X-RateLimit-Limit — the ceiling that applies to this request, in requests per minute (600 here is only an example of a configured limit).
• RateLimit-Remaining / X-RateLimit-Remaining — how many more requests you can make in the current window.
• RateLimit-Reset / X-RateLimit-Reset — when the window resets, as a Unix timestamp (seconds).

These are advisory — a healthy client paces itself off Remaining and only falls back to retry-on-429 as a safety net. When no per-org limit is configured, these headers are absent.

What happens at the ceiling

When you’ve consumed the budget, the next request returns:

HTTP/1.1 429 Too Many Requests
Content-Type: application/problem+json
Retry-After: 12
RateLimit-Limit: 600
RateLimit-Remaining: 0
RateLimit-Reset: 1747512000
X-RateLimit-Limit: 600
X-RateLimit-Remaining: 0
X-RateLimit-Reset: 1747512000

{
"type": "https://developers.sendops.dev/api-reference/errors/rate_limited",
"title": "Rate limit exceeded",
"status": 429,
"code": "rate_limited",
"detail": "You have exceeded the per-minute API key request limit; retry after the time indicated in Retry-After.",
"retry_after": 12
}

Retry-After is the canonical signal — sleep at least that many seconds before retrying. It is the number of seconds to wait; the Reset headers give the absolute Unix timestamp of the window reset, and the problem body’s retry_after mirrors Retry-After.

Handling 429 in code

The safe pattern is exponential backoff with jitter, honouring Retry-After as the minimum delay.

async function callSendOps(path) {
for (let attempt = 0; attempt < 4; attempt++) {
  const res = await fetch(`https://api.sendops.dev${path}`, {
    headers: { Authorization: `Bearer ${process.env.SENDOPS_API_KEY}` },
  })
  if (res.status !== 429) return res

  const retryAfter = Number(res.headers.get("Retry-After") ?? 1)
  const jitter = Math.random() * 250
  await new Promise(r => setTimeout(r, retryAfter * 1000 + jitter))
}
throw new Error("rate-limited after 4 attempts")
}

A few rules:

• Never retry faster than Retry-After. Doing so spends budget without making progress and slows your own recovery.
• Cap retries — runaway loops are the most common reason a service stays rate-limited indefinitely.
• Add jitter so multiple clients don’t lock-step into the same retry instant.

The Send API

The Send API (early access, on ingest.sendops.dev) applies the same per-minute limit as above, from the same budget as the key’s Public API requests, with the same headers and the same 429 rate_limited. Three things differ:

• A monthly fair-use limit. Once your plan’s monthly number of live API sends is used up, POST /v1/send answers 429 send_quota_exceeded with limit, used and resets_at. Its Retry-After is the time until the billing period resets, which can be days, so don’t retry on it. See Monthly fair-use limit.
• A cap while requests are queued. When SendOps is briefly unavailable and the Send API is answering 202 queued, each key may queue about 100 requests per 10 seconds; past that it gets 429 rate_limited with Retry-After: 10.
• Your SES account’s limits surface as 503. A send that hits your SES account’s send rate or daily quota is 503 upstream_unavailable with Retry-After, not a 429. Nothing was sent; retry with the same Idempotency-Key.

Auth-failure limits (IP-based)

Unauthenticated traffic and traffic from invalid keys is rate-limited per IP, with a tighter ceiling, so we don’t get DDoS’d by credential-stuffing or scanner probes. You should never see this limit in normal use — it only trips when you’re sending many requests with bad keys from the same source.

Test vs live quotas

Test keys run at 10% of the live limit. That’s intentional — test keys are for CI smoke checks and unit tests, not load tests. If you find yourself wanting to load-test against test keys, use live keys against a non-production SendOps account instead.

Tips for high-volume readers

Most readers won’t come close to the limit, but if you’re paginating large message lists or polling reports on a tight loop:

• Cache responses. Reports are computed from event data that updates within seconds — caching for 30–60 seconds on your side is usually safe and reduces request count dramatically.
• Use the largest limit you can. Lists support limit up to 200 on most endpoints. Fewer requests, same data.
• Filter server-side. Don’t pull all messages and filter client-side; use the available query parameters.
• Subscribe instead of poll where possible. For event-driven needs, configure a webhook in the dashboard rather than polling the API.