Reference

Rate limits & credits

Siftzo limits how many requests you can run at once, not how many you send per second. Credits limit how much you can use per month.

Limits per plan

Limits per plan
PlanConcurrent requestsCredits per monthExtra credits
Free21,000None
Starter1025,000$2.20 per 1,000
Growth25100,000$1.70 per 1,000
Scale50350,000$1.40 per 1,000

Concurrency

Concurrency is the number of requests that can be running for your account at the same time, across all your API keys. Sync requests count until they return; async jobs count from the moment they are queued until they finish.

When you are at the limit, a new request is refused right away with 429 concurrency_limit and a Retry-After: 1 header. It is not queued and not billed. Run a worker pool no larger than your limit, or retry after a short wait.

429 Too Many Requests
{
  "error": {
    "code": "concurrency_limit",
    "message": "concurrency limit of 10 reached"
  }
}

Credits

Credits per request
RequestCredits
HTTP request (render=never)1
Browser request (render=always, or auto when needed)10
SERP request (any engine, any tier)10
Browser API session1 per 10 seconds
Crawl API pagesame as a scrape of that page; robots.txt fetches are free
Batch itemsame as the scrape it runs
Markdown output (format=markdown)no extra credits
Data API recordsame as a scrape of the tier that fetched the page; the AI fallback adds extraction credits
AI extractionscrape credits + LLM tokens: ⌈(input × 2 + output × 10) / 300⌉ with the default model
Feedseach request in a run is billed like the API call
Proxy access traffic (in and out together)6 per MB
Failed request (blocked, timeout, proxy error)0

Before a request runs, Siftzo sets aside the most it can cost: 10 credits for render: "auto" and "always", 1 for "never", and 10 for a SERP request. When it finishes, the reservation is replaced by the real price. While requests are running, credits_used in GET /v1/usage includes these reservations.

Billing rules

  • You pay for the tier that produced the result, once per request, not per attempt. A SERP request costs the same on every tier.
  • A request is billed only when the target answered without being blocked, with status 2xx, 3xx, 404 or 410.
  • Other target statuses (for example 403 or 500) are returned to you but not billed.
  • Blocks, timeouts, unreachable targets, proxy errors and refused requests cost 0 credits.

Monthly limit and reset

Credits are counted per calendar month in UTC. On the 1st of each month at 00:00 UTC the count starts again from zero; unused credits do not roll over.

Your monthly limit is your plan's included credits plus the overage you allow. On a paid plan you set a spending limit for overage in the dashboard; on the Free plan the limit is the included credits.

A request that would take you over the limit is refused with 402 credits_exhausted. Because of the reservation above, an auto request needs 10 free credits to start, even if it ends up costing 1.

Credits belong to your organization. Its teams can each have a monthly cap of their own: a request with a team's key must fit both the organization's limit and the team's cap, and over the cap it gets 402 credits_exhausted with a message that names the team's cap. Owners and admins set caps under Teams in the dashboard.

Watch credits_remaining from GET /v1/usage to slow down before you hit the limit, or raise your spending limit from the dashboard.

Other limits

  • timeout_ms: 5,000 to 90,000 ms, default 60,000.
  • Request body to the API: 2 MB. Body forwarded to the target: 1 MB.
  • Up to 30 custom headers. URLs up to 2,048 characters.
  • Target responses up to 10 MB.
  • Async results are kept for 24 hours.

AI extraction and the Data API

  • timeout_ms: 5,000 to 50,000 ms, default 30,000, for the scrape. The language model has its own 50-second budget.
  • At most 30,000 input and 4,096 output tokens per model call, and at most 2 calls (one repair) per request.
  • A monthly LLM token cap per account, by default 5,000,000 tokens. Past it, extraction is refused with 402 llm_tokens_exhausted until the next month (UTC).
  • Schemas up to 16 KB, 5 levels deep and 100 properties. Prompts up to 4,000 characters. See AI extraction.

Feeds

  • 20 feeds per account; the 21st gets 409 feed_limit.
  • 1–100 URLs per feed. Schedules run at most once an hour; Run now at most every 15 minutes, one run at a time.
  • Run results are kept for 7 days. See Feeds.