Install local AI with Ollama on the PC
This guide walks through installing Ollama 0.4 on Windows 11 with 32 GB RAM for everyday use.
- ki
- produktivitaet
- digitalisierung
AI-assisted · editorially reviewed Details
On 17 August 2026 the installation of Ollama 0.4.1 runs on a Windows 11 machine with 32 GB RAM.
Check hardware
First the available VRAM is verified. With less than 8 GB VRAM this step is skipped, yet only models under 7B parameters then fit into memory. The 2025 pcgameshardware.de special on local AI names this value as the minimum requirement.
Download Ollama
The installer is fetched from ollama.com. The file size is 142 MB. Without the download no local model starts.
Run installation
After double-clicking the EXE a window with a progress bar appears. The process takes 90 seconds on a Ryzen 7 5800X. Skipping it prevents later command-line execution.
Pull model
The command ollama pull llama3.1:8b downloads the 4.7 GB model. On a 100 Mbit connection the transfer takes 8 minutes. Omitting this leaves the reply function empty.
Start test run
The command ollama run llama3.1:8b opens an interactive shell. A simple prompt such as “Summarize the latest project report” returns output after a few seconds – how fast depends on model size and hardware. Without the test the later integration into tools like Obsidian fails.
Integrate into daily work
Ollama connects to Microsoft To Do via the API on localhost:11434. Missing this link keeps generated texts isolated and requires manual copying.
Free website check
A separate scan of the company site at https://log-sysdev.de/website-check reveals load times and SEO issues that become easier to fix after the AI setup. The check is free and delivers a report within two minutes.
Common first-query error
Users often forget to set temperature to 0.7. The model then repeats phrases or produces empty sentences. This value is stored in the Modelfile and must be adjusted before the next run.