n8n 2.47 adds per API key rate limits, so one runaway workflow no longer blocks every team on the instance
One workflow looping over 1,000 items can send about 1,000 API calls at once. n8n finally gives admins a built in limit for that.
What happened
From version 2.47.0, n8n lets admins cap how many public API requests each API key can make. It is available on every plan, on n8n Cloud and self hosted, and it is off by default.
The n8n blog explains the problem it solves. On a shared instance, many clients and workflows call the public API, including workflows that call their own instance through the n8n node or the HTTP Request node. Because a node runs once per item, a list of 1,000 items sends about 1,000 requests at once, and that can block everyone else. Some customers had been building their own rate limiters around n8n.
How it works:
- Reads and writes get separate limits, globally and per endpoint, and each API key keeps its own count.
- Over the limit, n8n returns a 429 with Retry-After and rate limit headers.
- Changes in Settings, Policies, Public API Limits or through the API apply without a restart. Environment variables override both on self hosted.
- In multi main setups the count is shared through Redis.
It pairs with idempotency keys, available since 2.45.0. Send an Idempotency-Key header on a POST, and if it succeeds n8n stores the response for 12 hours, so a retry with the same key and body returns that response instead of running again.
My take
This is the kind of release that does not trend but saves a weekend. Two things I would do this week on any shared instance:
- Turn the limits on with a tighter cap on writes than on reads. Writes are where a loop does damage.
- Fix workflows that call their own instance. n8n says those nodes fail on a 429, and recommends Batching over Retry On Fail, because Retry On Fail waits at most 5 seconds and often retries inside the same window. Their example: five items a second is 300 requests a minute, below every default limit.
Then add idempotency keys to any client that creates things through the API. A retry that creates a duplicate record is worse than a retry that waits.
One limitation to know: every key on an instance shares the same limits. If one integration needs more headroom, n8n's answer is a separate instance.
More posts
- LangSmith can now hide trace inputs and outputs from engineers while still showing them latency, cost and run treesNov 19, 2026
- Microsoft-Decision-1.1 is generally available for routing and classification at $0.042 per million input tokens, with free outputNov 19, 2026
- LangSmith now ties coding agent spend to pull requests, so teams can see cost per merged PR across Claude Code, Codex and CursorNov 19, 2026
