What n8n actually costs: self-hosted vs cloud
n8n Cloud versus self-hosting, with the break-even worked through - including the maintenance hours nobody costs and when n8n stops being the right tool.
The tools we build with, what they really cost, and when we would not use them.
Three notes on tools we use in production. Self-hosted n8n is free for internal use but, in our estimate, costs $14-150 a month in infrastructure plus real maintenance time. Below about a million vectors, pgvector in the Postgres you already run is enough. Use LangGraph when a workflow needs durable state, and write a retrieval pipeline or short agent loop yourself. Each page gives the numbers and says when we would choose something else.
n8n Cloud versus self-hosting, with the break-even worked through - including the maintenance hours nobody costs and when n8n stops being the right tool.
What each vector database actually costs at 100k, 1M and 10M vectors, what it costs to operate, and why pgvector is enough more often than vendors suggest.
LangChain and LangGraph earn their place in some situations and get in the way in others. Where the abstraction helps, where it costs you, and what we use.
Dated articles live on the blog.
A free strategy session: what is buildable, what it costs, what we would do first.