How Webflow Powers Agentic Workflows With n8n

Webflow powers tens of millions of websites, and giving customers one place to bring all their systems together is critical. "Type a sentence. Get a working agent" is the pitch we shipped, and n8n is the engine running underneath it.

Here's the twist: nobody at Webflow builds these workflows by hand. An LLM does, writing workflows at speed, for everyone, all the time. We treat n8n's workflow JSON as a compile target, not something anyone edits directly. The model writes against a typed SDK, we run it through n8n's own structural validator plus an allowlist of trusted node types, and failures get fed straight back into the next generation attempt instead of surfacing to a person.

Multi-tenancy runs on n8n's Projects, one per tenant, with traffic routed across hot and cold n8n instances, hot ones handling live runs, cold ones absorbing background noise. 

We'll show you exactly where it broke, and what we built to hold it together.

• Understand how Webflow generates working n8n agents directly from natural language, with no one hand-building workflows
• See how n8n's workflow JSON is treated as a compile target, written by an LLM against a typed SDK, not edited by hand
• Learn how generated workflows are validated: n8n's structural validator, an allowlist of trusted node types, and a feedback loop that retries failures automatically instead of surfacing them to a person
• Discover how multi-tenancy is handled at scale using n8n's Projects, one per tenant
• Identify how traffic is routed across hot and cold n8n instances to separate live runs from background load