For client work, I would lean toward one production instance per customer when isolation, handoff, and client ownership matter. A single centralized instance is easier at the beginning, but the maintenance risk grows quickly: shared credentials, unclear boundaries, harder client offboarding, and more anxiety around giving customer visibility.
The model I like is:
- Agency sandbox for templates, experiments, and reusable patterns.
- Client-owned production instance for live workflows.
- A release checklist when moving from sandbox to production.
- A central inventory outside n8n that tracks client, workflow, owner, health checks, issues, and reports.
That last point is important. Instance structure solves access and isolation, but it does not automatically solve maintenance. You still need to know which workflow belongs to which client, what “healthy” means, when it last ran, what broke, what was fixed, and what the client needs to know.
That is also the space Maintain Flow is focused on for agencies: a post-launch operating layer for checks, check runs, issues, resolutions, and client-ready reports.