AI THAT NEVER LEAVES HOME
UtilitiesOllama — run every AI feature on your own machine
Connect Drazil to Ollama in four steps, and chat, notes, diagrams, the LaTeX assistant — everything with an AI badge — answers from your own hardware.
Opens locally in your browser. The connection stays on your machine.
Ollama is a free, open-source runner for large language models on your own computer. Connect it to Drazil and the arrangement changes fundamentally: every AI feature in the app — chat, notes, transcript tools, diagrams, simulations, the SVG generator, the LaTeX assistant, anything with an AI badge — runs against the model on your machine. Nothing about your prompts or your material leaves home.
The connection is four steps, in the order the app expects them: install Ollama from its official site; pull a model (the guide suggests sizes that suit typical laptops); set the OLLAMA_ORIGINS environment variable so the server accepts requests from Drazil's origin; restart, then choose Ollama as your provider in settings. The origin allowlist exists so a random web page cannot quietly drive your local model — Drazil has to be explicitly permitted, which is good manners on the internet.
Model choice is the one decision with texture. Small models — roughly 3 to 8 billion parameters — run well on a modern laptop with 16 GB of RAM and handle summaries and rewriting comfortably. Larger models reason better and ask more of your hardware. A quantized 7B or 8B build is the sweet spot for most machines.
This is the strongest privacy posture Drazil offers — no third party in the loop at all, no per-token billing, no rate limits beyond your own computing. If the laptop model someday feels small, the same settings screen points Drazil at a bigger machine on your own network.
Local means private, not infallible — small models still invent facts with confidence, so keep the verify-what-matters habit.
What you can do
The parts people actually reach for.
Four-step connection
Install, pull a model, set OLLAMA_ORIGINS, restart — done.
Every AI feature, local
Chat, notes, diagrams, the LaTeX assistant — all on your machine.
Model size guidance
From 3B quick responders to larger builds, matched to your hardware.
No keys, no billing
No API keys involved; no per-token costs, ever.
Room to grow
Point Drazil at a bigger machine on your own network later.
The strongest privacy posture
No third party in the loop — the whole arrangement is yours.
How it works
Three steps, no detours.
Install and pull
Install Ollama, then pull a model suited to your machine.
Allow the origin
Set OLLAMA_ORIGINS so Drazil is permitted to talk to it, then restart.
Choose the provider
Pick Ollama and your model in Drazil's AI settings — and every badge goes local.
Questions, answered
The ones people ask first.
What if it will not connect?
Check OLLAMA_ORIGINS first, the model name second, and restart after setting the variable — those two cover almost every case.
Is a local model as good as a hosted one?
Smaller models trade breadth for privacy; a quantized 7B–8B build is the sweet spot for most laptops.
What does this cost?
Nothing. Ollama is free and open-source, and your hardware is the only meter.
More in Drazil
One workspace, 44 tools — all free, all local.
Opens locally in your browser. The connection stays on your machine. · or browse all 44 tools