If an AI agent or script calls your n8n instance through the public API, update to n8n 2.47.0 and turn on per-key rate limits. That way one runaway loop cannot freeze every other automation. For any call that creates records, such as leads, invoices or orders, send an idempotency key so a retry never produces a duplicate.
n8n added limits per API key on every plan, both Cloud and self-hosted, with separate caps for reads and writes. A client that hits its cap gets a 429 response with a Retry-After header saying when to try again. Idempotency keys have been available since 2.45.0. When a POST succeeds, n8n stores the response for 12 hours, and a retry with the same key and body gets that stored response instead of creating a second record.
What this changes for a small business
A setup I see often is one instance, a Claude agent creating tasks or CRM records through the API, and workflows calling their own instance. A node runs once per item, so a list of 1,000 contacts means about 1,000 requests in one burst. Until now, teams solved this with a homemade limiter. Now it is a setting.
One caveat: limits are off by default, and the n8n node and HTTP Request node simply fail on a 429. You need to add Batching to those nodes. n8n's example is 5 items per second, or 300 requests a minute, which stays below every default limit. Retry On Fail does not solve this, because it waits at most 5 seconds and often retries inside the same window.
Why it matters: duplicate records and a frozen instance are the two failures owners notice first. Turn on the limits, add batching to bulk nodes, and give each agent its own API key so you can see which one is misbehaving.
Source: n8n Blog