Using n8n to Power a Scalable SaaS for AI Assistants – Self-Hosted or Cloud?

Built with n8n
,
1

Hi everyone,

I’m working on launching a new multi-modal AI Assistants SaaS service and I’m exploring using n8n as the workflow engine behind it. My goal is to create a generic workflow that can be leveraged by each of my SaaS users to automate various tasks (e.g., connecting to APIs, orchestrating AI models, etc.).

Initially, I’m expecting about 10,000 users within the first six months, but I plan (and hope!) to scale beyond 100,000 users as the product grows. I’d love to get some insights from the community on the following questions:

  1. Hosting Options
  • Would n8n’s cloud/hosted version be sufficient for this level of usage, or would it be better to self-host n8n in my own infrastructure for more control over resources and scaling?
  • If self-hosting, are there recommended best practices or existing guides for clustering n8n to handle potentially large volumes of concurrent workflow executions?
  1. Scalability
  • Is n8n’s architecture built to handle the kind of load I might see from tens of thousands (eventually 100k+) of active users?
  • Are there any known limitations or bottlenecks to be aware of when running n8n at scale (e.g., concurrency limits, database constraints, etc.)?
  1. Performance Tips
  • For those who’ve scaled n8n to a large user base, do you have any tips or configurations that helped ensure smooth performance (e.g., caching strategies, job queuing, or horizontal scaling setups)?
  1. Implementation Approach
  • Has anyone used n8n in a multi-tenant SaaS context before? Any suggestions or potential pitfalls when creating a “generic workflow” template that each SaaS user can customize?
  • Would you recommend separate n8n instances per customer, or a single, more robust instance that handles all customers with strict resource isolation?

I’d appreciate any experiences, insights, or guidance you can share. I’m particularly focused on ensuring I choose the right hosting approach (cloud vs. self-hosted) and understanding how to best scale n8n to handle many users and potentially large workflow volumes.

Thanks in advance for your help, and I look forward to hearing your thoughts!

2

I recommend to check the n8n “Sustainable Use” license. You might have to pay a significant chunk of money per year for using n8n inside SaaS offering if you want to allow your customers to enter their credentials for various services.

3

Hey @BillyBot

As @Anthony mentioned there is a very good chance based on your description that you will need an enterprise or embed license for what you are doing, I would start by booking in a call to chat with the sales team and they will be be able to go over your questions at the same time, however I have put some quick answers below.

  1. I would recommend self hosting n8n and looking into queue mode, we have all of the information in the docs including some benchmarking info.

  2. Same as above really, assuming your database and redis server is configured to handle it you can add as many webhook workers and normal workers as you need and let your load balancer route the traffic.

  3. Assuming your database and redis is configured to handle the load you should be fine.

  4. n8n itself isn’t designed as a multi-tenant application so you would need to work out a way to handle this, what we do with our cloud solution is every user gets their own n8n container instance so there is no risk or it is at least extremely unlikely that data is going to be shared between instances.

4

I was planning on doing the same. Are you-- advisors- saying that if one is to run a multi-user experience, that requires authentication and has the potential of gathering thousands of users, n8n may not be able to handle it?

Once thing is to pay for it, of couse, the other thing is not be able to handle it. Can you share more of your thoughts on this please?

Thanks.

5

I’d suggest n8n cloud is good to start with to build the foundation of your idea; and if your service gets bigger, you can shift to a self-hosted model then.