Skip to content

ThorOS 0.4 — first public release

Choose a tag to compare

View all tags
@csakytech csakytech released this 26 Jun 18:12
· 35 commits to main since this release
v0.4.0
93a8a40

ThorOS 0.4 — AI-first, local-first, voice-driven Linux

ThorOS is an operating system built around a local AI assistant. A voice assistant — "Jarvis" by default, rename it to anything — drives your desktop, opens and controls apps, writes documents, and answers questions. 100% local: no cloud, no account, no telemetry. Built on Debian 13.

⬇ Download

ThorOS-0.4-amd64.iso  ·  4.3 GB

SHA-256: bb801780a249f5523007c6832494a19f4ab6dbd426e434d16e6b7ad273171984

Verify after downloading: sha256sum ThorOS-0.4-amd64.iso

Install

  1. Write the ISO to an 8 GB+ USB stick with balenaEtcher or Rufus.
  2. Boot from it and choose Start installer (or Live system to try it without installing).
  3. On first boot it auto-detects your GPU and downloads the local AI model (~5 GB — needs internet, takes a few minutes). The assistant is ready once that finishes.
  4. Say your assistant's name (default "Jarvis") and start talking — or open the text-chat window.

➡️ New to Linux? Read the 2-minute Getting Started guide.

First steps & troubleshooting

  • Be patient on first boot. ThorOS spends ~10–20 minutes (one time, needs internet) downloading its AI model. Jarvis says "I'm still getting set up" until that finishes — that's expected, not a bug.
  • Talk to it: say "Jarvis" then your request — "Jarvis, open Firefox", "Jarvis, write a thank-you note", "Jarvis, what's 15% of 240?". No microphone? Use the Jarvis Chat window.
  • Rename it: say "Jarvis, call yourself Athena" — then just say "Athena." The name is the wake word (no "hey").
  • Replies slow? On a PC with an NVIDIA card, the fast GPU driver finishes installing on the first reboot — restart once and it speeds up a lot.
  • Check the AI downloaded: open a Terminal and run ollama list — you should see qwen3:8b. If the list is empty, download it manually with ollama pull qwen3:8b and wait for 100%.

Notes for this release

  • Hardware-agnostic: boots on any 64-bit PC. An NVIDIA GPU is auto-detected and used for fast responses; it also runs CPU-only (slower for the 8B model).
  • Network firmware + SSH are included; the installer configures networking with DHCP.
  • Internet is needed once on first boot to fetch the AI model; fully offline-capable afterward.

Requirements

  • 64-bit PC · 8 GB+ RAM (16 GB recommended) · ~25 GB disk
  • Microphone + speakers for voice (or use the text-chat window)

🌳 Free and open source · yggdrasilai.org · Support the project

Assets 2
Loading