Local AI

Language models,
running at home.

Two dedicated servers run open-source language models for the whole family. Each person has their own isolated chat instance — no data shared with any cloud provider, no query leaving the house.

Ollama & Open WebUI

Two servers,
one architecture.

// PRIMARY

The first AI server runs Ollama as the language-model backend with an AMD Radeon GPU carrying 32 GB of VRAM. Open WebUI provides the chat interface. With 64 GB of system RAM, several large models can stay loaded at once without swapping.

// SECONDARY

The second AI server is identical in architecture — Ollama and Open WebUI on a matching AMD GPU with 32 GB of VRAM. It serves the rest of the family, with each person's chat history kept separate from everyone else's.

Open-source models

Capable models,
fully local.

The servers run current open-weight language models in the 30-billion-parameter range, quantised to fit comfortably in 32 GB of VRAM. Different models are kept available for different strengths — general conversation, coding, and reasoning — and can be switched per chat.

Files can be dropped directly into a conversation: a document attached to a chat is embedded locally and used as context for that conversation, with a local embedding model handling the retrieval. Nothing about the file or the conversation is sent anywhere outside the house.

Chrome Extension

Translate the web
with your own model.

Local LLM Translator

Whole-page translation, nothing sent to the cloud

This custom Chrome extension translates entire webpages using the local language-model instance running at home. No text is sent to any cloud service. The model is selected from a dropdown loaded directly from the local instance, and translations stream in word by word as the model generates them.

Visited pages are cached — revisiting a translated page reloads instantly without a second inference call. Domains can be marked for automatic translation on every visit.

Chrome Extension Streaming Translation cache Local model
→ Chrome Web Store
Beyond chat

The same models,
powering everything else.

These two servers are not just for chatting. The same local models drive the other AI projects on this site: the podcast generator's script writing, the transit app's status briefings, the shopping app's product matching, the WhatsApp agent's replies, and the AI-driven Ansible management. One local language-model backend, many uses — all without a cloud subscription.

Ollama Open WebUI Local embeddings Per-user isolation