No Docker - Self Hosted AI Starter Kit Help?

Questions
1

Referring to an earlier thread on the Self Hosted AI Starter Kit without using Docker…

https://community.n8n.io/t/self-hosted-ai-starter-kit/56525

Has anyone come up with an actual procedure for doing this, or at least some notes on how they’ve done it?

The reason I ask is that using the “Docker” solution does not take advantage of a GPU. It only uses CPU and “Native” RAM.

@ThatPrivacyGuy

It looks like you've done this but I'd love more detail if possible.

Thanks,

Scott
NN4W

2

It looks like your topic is missing some important information. Could you provide the following if applicable.

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
3

Hey @NN4W

Welcome to the community :raised_hands:

The self hosted AI stater kit has docker profiles to work with the GPU as well, did you run into issues when trying the GPU profile?

4

Hi Jon,

Thanks for the reply!

Unfortunatly yes… The GPU profile made no difference.

The Docker versions seem to work on a Windows based OS host, but not on Linux.

I should have given more information on my configuration.

I’m running an MX Linux host, which is actually Debian 6.1.115-1.

The GPU is an Nvidia Geforce 3090.

Running all apps natively, (Ollama, OpenWebUI and N8N), work fine and use the GPU.

Using the “Starter Kit” with Docker, all apps default to use the CPU, ignoring the GPU altogether.

I ran across a bug stating the same but I don’t have the reference handy at the moment. I’ll dig into it again and see if I can find the bug reference.

I have no problem switching from a Debian based Linux Distro if an Arch or RHEL version works, I just do not want to run Windows.

Thanks again,

Scott
NN4W

5

I have some of the components talking to each other all installed locally without Docker.

I’m still working on some of the details.

I’d really appreciate some input from anyone going this route.

Adding Docker into this just adds another layer of complexity that absolutely is not needed.

Any input wouldbe greatly appreciated.

6

As a reference, please look at: