JigsawStack
AI · AI
Web search, AI scraping, speech to text, translation, and OCR on their own key.
Connecting JigsawStack in practice
JigsawStack applies one flat limit across every model and endpoint — 50 requests per second per project — and reports it on each response through x-jigsaw-rate-limit-limit, -remaining and -reset, where the reset value is a Unix timestamp in milliseconds rather than a delay. The limit is per project, so issuing extra keys buys no extra throughput.
Rate limits
- 50 requests per second per project, across all endpoints and models.
- 429 above the limit; higher limits are arranged with support rather than by changing plan.
- Headers: x-jigsaw-rate-limit-limit, x-jigsaw-rate-limit-remaining, x-jigsaw-rate-limit-reset (Unix timestamp in milliseconds).
Worth knowing
- x-jigsaw-rate-limit-reset is an absolute Unix timestamp in milliseconds, so treating it as a number of seconds to sleep overshoots by three orders of magnitude.
- Because the budget is per project rather than per key, splitting agents across keys does nothing — separate projects are the only way to isolate throughput.
- One limit covers cheap calls and expensive model inference alike, so the rate ceiling and the cost ceiling are entirely different conversations.
Checked against Rate limits (2026-09-29), Documentation index (2026-09-29)
What an agent can do
Each action is granted on its own. An agent allowed to read is not thereby allowed to write, and the scope beside each row is what the acting user must have connected for it to run at all.
Often connected alongside
Put JigsawStack behind one governed endpoint.
Same permissions, same audit trail, whatever else you connect next.