Local AI Stack Installer
Ollama, Open WebUI, embeddings and a working RAG knowledge base on your own GPU box — installed and wired together by one script.
What it does
Drivers checked, packages installed, services enabled, ports verified.
Installed, connected and confirmed talking to each other.
The right model, with the full tag, so retrieval actually works.
A knowledge collection created and attached to a model you can use immediately.
Proper systemd units with lingering enabled, not a terminal you have to leave open.
A diagnostic that tests provider, endpoint, model, chat and embeddings in sequence.
Every part of a local AI stack is documented somewhere. Getting all of them talking to each other, surviving a reboot, and actually answering questions from your own documents is the part nobody writes down.
This installer does the whole thing: GPU drivers checked, Ollama installed and pulled, Open WebUI running as a proper service that survives logout, an embedding model wired up, and a knowledge collection ready to take your documents. It also sets up the systemd lingering that otherwise silently kills the service the moment you close your session.
Specifications
| Runs on | Ubuntu 22.04+ / Debian 12+ |
|---|---|
| GPU | NVIDIA recommended, CPU supported |
| Installs | Ollama, Open WebUI, embedding model, systemd units |
| Time | Around 20 minutes on a clean box |
Questions about this product
What if I already have Ollama?
It detects an existing install and wires around it rather than replacing it.
Does it need a GPU?
No, but CPU-only inference is considerably slower.