DEV Community

News
News

Posted on Edited on

n8n adds per-key rate limits for the public API

Infographic cover

๐Ÿ“… November 13, 2026 โ€ข Source: n8n Blog

n8n has added per-key rate limits for the public API, per the n8n blog and official docs. From version 2.47.0, admins on every plan โ€” n8n Cloud and self-hosted โ€” can cap how many API requests each individual API key can make. The feature ships in all self-hosted editions as well, it is off by default, and the instance owner or admin turns it on. The documented settings schema spans four limit families โ€” global read, global write, per-endpoint read and per-endpoint write.

The mechanics, per n8n docs: limits are counted per API key, with separate settings for reads and writes, both across all endpoints and per endpoint. Two keys for the same user count separately. Each request counts toward two limits at once โ€” global and endpoint โ€” and response headers show the limit with the fewest requests left. An over-limit request gets a 429 with Retry-After and rate-limit headers, and a rejected request does not count toward any limit and is never run. Response headers include X-RateLimit-Limit, Remaining and Reset.

Defaults exist for every limit, and changes via the settings page or API apply immediately without a restart, per the docs. On multi-main setups, every main shares one count per API key through Redis, so a keyed-out client gets a 429 from every main until the window ends. The motivation, per the n8n blog: workflow nodes run once per item, so a 1,000-item list can send about 1,000 requests at once, and some customers had built their own rate limiters around n8n. The docs show a Retry-After example of 18 seconds against a limit of 600.

One nuance from the docs matters for automation builders: n8n's own workflow nodes that call the instance API, like the n8n node and HTTP Request node, do not automatically honor Retry-After โ€” the node fails on a 429. The documented mitigation is Batching in the node plus an Idempotency-Key header for safe retries. Rate limiting also pairs with idempotency keys, available since n8n 2.45.0, so a repeated call does not double-apply a change. Honoring Retry-After sits with the calling client, per the same docs.

Key figures

IMPACT: teams sharing one n8n instance stop blocking each other โ€” a runaway workflow now hits its own key's limit instead of starving every other team on the instance. In practice, give each team or agent its own API key, set read and write caps per key, have internal workflow calls use Batching, respect Retry-After on external clients and attach idempotency keys to retried calls. Self-hosted operators get parity with Cloud, since the same controls work on every plan from 2.47.0.

Sources:
https://blog.n8n.io/new-in-n8n-public-api-rate-limits-that-keep-every-team-on-your-instance-running/
https://docs.n8n.io/connect/n8n-api/rate-limits
https://docs.n8n.io/connect/n8n-api/instance-settings/settings-rate-limits

IMPACT checklist


Continue the conversation: https://t.me/AIAutomationArchitecture ยท Bluesky

AI Automation

Top comments (1)

Collapse
 
aiautomationplaybook profile image
AI Automation Playbook •

Useful write-up โ€” the part that bites in production is the node failing on a 429 instead of honoring Retry-After. Two things from running approval-gated workflows: (1) test the 429 branch in a mock run before shipping โ€” a workflow dying mid-batch silently drops the rest, and mock E2E is where I caught that class of bug (a Stripe leg that was never connected, a Join running twice). (2) If the 429'd call sits behind an approval step, the approval is spent but the action never happened โ€” tie the idempotency key to the approval record so the retry doesn't double-apply.