ThorOS 1.0 — the first stable release

Warning

The 1.0 ISO has two critical first-boot bugs — fixed in ThorOS 1.1, the current release.

• Setting a root password during install leaves your user without admin (sudo) rights — and unable to receive self-updates.
• NVIDIA machines silently stay on CPU forever (the driver installs but is never activated).

New installs: download the 1.1 ISO instead. Already installed 1.0? The two-minute repair is in the 1.1 release notes.

ThorOS 1.0 — the AI-first, local-first, voice-driven operating system. Talk to your computer; it does the work. This is the first stable release.

⬇ Download the ISO (4.6 GB): https://www.yggdrasilai.org/downloads/ThorOS-1.0-amd64.iso
sha256: 8de31ee9d4c726c200711a5ddfc3149a456a30c07581c4b2622f76f0e63973fb

(The ISO is hosted on the project site — GitHub's 2 GB asset limit can't hold it. Write it to a USB stick and boot it to try ThorOS live before installing.)

---

Talk to your computer in plain language and it gets real work done — opening apps, writing and editing documents, browsing the web, running commands, building software — entirely on your own machine. Private by default, your choice of local or cloud. Built on Debian, open source, and designed for everyone, including people who can't use a keyboard or mouse.

What ThorOS does

• Talk to it. Say the wake word and speak. Wake-word → speech-to-text → an on-device planner works out what you mean → specialized agents carry it out → it replies in a natural voice. All of it runs on your hardware; unplug the internet and it still works.
• It does real work, not just chat. Agents open programs, create and edit documents (bold, lists, find-and-replace, export to PDF), manage files (always confirming before it deletes), run terminal commands, search the web, set reminders, and give you spoken briefings.
• Browse the web by voice. Say "click" to number every link and button on the page, then "select 4" to open one — or "read the page" to have it read aloud. Real hands-free browsing.
• Smart Help — "what can I say here?" Say "Jarvis, help" anywhere and a card shows the exact commands that work right where you are. Every command is numbered — just say "do number 3" to run it.
• Development Mode. "Jarvis, enter development mode," describe what you want to build in your own words, answer a few questions, and AI agents scaffold and build runnable software with you.
• It remembers. "What was I working on yesterday?" — ThorOS keeps a private, on-device activity journal and can recap your day.
• Many models, your choice. Bind different local LLMs to different jobs (planner, coder, writer, reasoner); run several at once if your machine can handle it.
• It explains itself. Ask "why did you do that?" and it tells you the real reasoning behind its last action.
• Built for every generation. Hands-free login, spoken interaction, and a goal-oriented design so the computer works like a helpful person — not a machine you have to fight with.

Private by default

The speech recognition, the language model, and the voice all run locally. Your documents, your commands, and your conversations never leave the machine unless you choose a cloud model. No tracking.

Already running ThorOS?

It updates itself — you'll be offered 1.0 automatically, or just say "update".

Under the hood

Debian + GNOME, a permissioned multi-agent runtime, Whisper (speech-to-text), Ollama (local LLMs), and Piper (voice) — with a first-boot that detects your GPU, picks a model tier, and starts the assistant. Free and open source under Apache-2.0 / MIT.

Powered by Yggdrasil AI · https://www.yggdrasilai.org