JARVIS Omega v1.2.0

JARVIS Omega v1.2.0

A full interface rebuild plus two neural voice engines and a new AI provider.

Install

Download JarvisOmega-Setup-v1.2.0.exe below and run it. Windows 10/11, 64-bit.
No Python needed — everything is bundled.

Windows SmartScreen will warn on first run because the installer isn't code-signed.
Choose More info → Run anyway.

---

New voices

Jarvis no longer has to sound like a 2005 screen reader. Settings → Voice now offers
three engines:

• Edge — Microsoft's neural voices. Free, no API key, 10 voices. Needs internet.
• Kokoro — runs entirely on your machine. 9 voices, fully offline after a one-off
  ~350MB model download (Settings → Voice → Download model).
• Windows — the original SAPI5 engine, kept as an always-available fallback.

If a network blip hits Edge, or the Kokoro model is missing, Jarvis falls back to
Windows SAPI rather than going silent.

GLM support

New AI provider alongside Ollama and OpenAI. Works with a Z.ai Coding Plan
subscription — set it up in Settings → AI Brain with Test connection before
committing. Supports glm-5.2, glm-5-turbo and glm-4.7.

Your API key is stored in a local .env file, never in config.yaml.

Redesigned interface

Everything was rebuilt around one design language:

• HUD — rendered with Pillow instead of raw Canvas, so the arcs are antialiased
  and the glow is real. Radar sweep, segmented rotating arc, counter-rotating ring,
  targeting brackets and pulse rings that react to state.
• Splash screen — a real boot sequence with a working progress indicator.
• Right-click menu, pop-out panels, editors and Settings — all rebuilt.
• All 18 themes still work and now drive the accent colour throughout.

Fixes

• Themes had no effect on the HUD. Open-mic mode sits in listening almost
  permanently, and the state colour was overriding the theme accent, so every theme
  rendered green.
• Clicking a HUD button teleported the window. The click handlers returned
  without updating the drag origin, so the next mouse movement jumped the window by
  the stale offset.
• Dragging was laggy — the renderer was competing with the window move for the
  frame budget.
• Opening the right-click menu broke scrolling everywhere. It bound the
  mousewheel application-wide and never released it.
• Update checks always failed. They pointed at a private repository, which
  returns 404 to anonymous clients. Releases now live in this public repo.
• Editors could save malformed YAML, silently breaking your commands on next
  start. Saves are now validated first.
• Settings' Close button floated beside the content instead of docking to the bottom.

Notes

• Auto-update reads this repository's release feed. Existing installs pointing at
  the old private repo migrate automatically on first launch.
• The app folder is ~455MB, mostly the bundled Whisper speech model.
• Kokoro's voice model is not bundled — download it in-app when you want it.