The owned automation stack

All three tools here have a genuine free tier, so a working configuration costs nothing while you evaluate it. The 03 COSTS table above breaks down what each vendor publishes. Three meters climb with use: n8n's execution counts and, more to the point, the infrastructure and engineering attention of self-hosting it; Relevance AI's usage credits, its model billing kept visible by your own keys; and Airtable AI's credits, worth watching early as agent-heavy work ramps. The engineering ownership dwarfs the licences here, so budget the people before the software.

No. n8n is the centre of gravity: self-hosted workflows with code-level control earn the stack on their own, and many teams run on it alone for a while. The numbered steps map the architecture, not the buying order. Adopt n8n first; add Airtable once flows need a structured system of record with approvals attached; bring in Relevance AI last, when judgement-shaped work fixed flows cannot express appears.

Airtable already gives you a structured system of record, and this stack keeps it as the anchored data layer, not a tool to replace. What you are graduating from is Airtable-alone. n8n adds owned, code-level workflows with no per-task meter on your own infrastructure; Relevance AI adds agents for the judgement-heavy work a database was never meant to do on its own. The records stay; the automation grows up around them.

n8n and Relevance AI both automate work, the one real overlap here. The rule is whether the task can be written down. n8n wins when the steps are deterministic and repeatable: a fixed pipeline you version and run without a model in the loop. Relevance AI wins when the work needs judgement fixed flows cannot express: the cases a rule would miss. They connect, so the boundary is a handoff, not a wall.

Often, and this tier says so plainly. Owning the automation only repays its bill with the engineering capacity to run self-hosted infrastructure and a real reason to: hard data-residency rules, or volumes where a per-task meter would bite. Without an engineer to keep n8n healthy, or where hosted per-operation pricing covers the load, ownership costs more than it returns. A team reaching for this on principle rather than need wants the competitive tier, the professional automation stack.

n8n runs on infrastructure you control, so workflow data stays put; the hosted members set the floor. Put confidential records into Airtable AI only under its enterprise terms, where admin controls govern the AI features, and grant Relevance AI's agents credentials deliberately: they act across connected tools, and an agent with edit access to the wrong table is an operational risk before it is an AI one. Decide which records and prompts cross into the hosted platforms before the fleet scales, not after.