Skip to content

Gnosis 1.3.0

Choose a tag to compare

View all tags
@DOMCHURCH DOMCHURCH released this 31 Aug 16:30
· 36 commits to master since this release
v1.3.0
eb7370d

Terms of Use, voice on by default, and updates that actually install.

Install

irm https://raw.githubusercontent.com/DOMCHURCH/Gnosis/master/scripts/install.ps1 | iex

Updates now actually land

autoInstallOnAppQuit was false, which does not mean "ask first" — it means discard. Gnosis downloaded each update, showed a toast, and if you closed the app without clicking Restart it threw the build away and re-downloaded 120 MB next launch to delete it again. Anyone who never clicked that button was pinned on their original install forever.

A staged update now installs when you quit. Nothing yanks a running agent out from under you — only your Restart click restarts a live app. It also re-checks every 6 hours instead of once at launch.

Terms of Use, and a record that proves it

New TERMS.md, shown on first run and after feature releases: what the agent can do, that its safeguards are best-effort rather than guarantees, that you're responsible for what you instruct it to do, no warranty, limited liability.

Acceptance is recorded in ~/.dom/acceptance.json — keyed on a sha256 of the exact text you were shown, not a version number, so editing the document by one byte re-prompts rather than leaving you recorded as agreeing to words you never saw. Append-only, with a copy of the accepted text so the checksum stays verifiable.

Nothing about it is sent anywhere. Recording acceptance on a server would mean collecting data about every install — exactly what the terms promise not to do.

Voice is on by default

Previously off and invisible unless you happened to know the wake phrase. Now on — but nothing listens until after the welcome window has told you it's live, what's sent where, and given you a switch to turn it off without leaving that window.

The window also offers the one-time openWakeWord + Kokoro download (~350 MB, free, local) when they aren't installed, since voice being "on" means nothing if the Python side was never set up.

Honest about the trade: wake-word detection runs entirely on your machine and no audio leaves it while it's just listening. Only what you say after "hey jarvis" is recorded and sent to Groq for transcription. Don't leave it on where others could be recorded without consent.

Also

  • Whisper detection: the speech gate was a fixed threshold that discarded anything quieter than normal speech. It's now measured from your actual room noise, and Chrome's noise suppression — which erases whispers — is off.
  • Apps launch by their real name: "MSI Afterburner" resolves through Start Menu shortcuts, and the launch wait went from 5s to 30s (measured: ~9s for Afterburner).
  • A model that falls into a repetition loop is cut off instead of emitting the same sentence 150 times.
Assets 5
Loading