MIT · open source · Windows 10/11

Talk to your PC.
It types for you.

Hold a button, speak, release — your words appear at the cursor in whatever app has focus. Or open a voice conversation that can actually do things: search the web, open your apps, run tasks on your machine. No subscription, bring your own OpenAI key.

Local wake words — no audio leaves the device while listening 15 languages, right-to-left included Key in Windows Credential Manager — never in a file
jarvis-demo.mp4 — 88s, real session, no mockup
What it does

Dictation that lands where you're typing.
An assistant that can act.

Two engines behind one hotkey. Economy transcribes with gpt-transcribe; Live holds a conversation on gpt-live-1. You pick per mode, and the app enforces a spend ceiling you set.

Dictation into any app

Focused field, any program. It never presses Enter, it refuses password fields, and it holds text for review in terminals.

Voice conversation

A GPT-Live session that can search the web and drive the machine through an approval-gated tool layer — nothing runs without your yes.

Local wake words

"Hey Jarvis" is detected on your machine by sherpa-onnx. Opt-in, off by default, and nothing streams while it listens.

The floating pill

A non-activating overlay with six states — idle, listening, thinking, speaking. It never steals focus from what you're doing.

Memory that helps

Local SQLite history plus vocabulary hints that measurably improve accuracy on your own names and jargon.

Private by construction

No telemetry, no analytics, no account, no server of ours, no localhost port. The key lives in Windows Credential Manager, DPAPI-wrapped.

Muscle memory

Five controls, all rebindable.

Every one can be moved to a different key, a different mouse button, or nothing at all in the setup wizard.

Mouse side buttonHold to dictate into whichever app has focus
F8Start and stop dictation, hands-free
Ctrl+Alt+SpaceOpen a voice conversation
Ctrl+Alt+PauseEmergency stop — ends the session and any running tool
EscCancel whatever is happening right now
What it costs

No subscription. You pay OpenAI, at cost.

Twenty minutes of dictation a day, five days a week, is about 45 cents a month on Economy. The app enforces your own spend ceiling and shows a running meter.

$0.0045/minEconomy dictation on gpt-transcribe — the default, and the most accurate
$0.05/minLive voice conversation on gpt-live-1, when you want a back-and-forth
$0Subscription, markup, account, or telemetry. Free and MIT-licensed forever.
Install

One line in PowerShell.

It finds your Python, clones Jarvis into %LOCALAPPDATA%\Jarvis\app, builds the virtual environment, installs the pinned dependencies, runs a smoke check and adds a Start Menu shortcut. Per-user and reversible: it never elevates, never installs Python for you, and never touches your execution policy or Defender settings.

Full install notes
  • Windows 10 or 11, Python 3.11+
  • Your own OpenAI key — entered in the app, stored in Credential Manager
  • macOS and Linux exit with a clear message today; a port is on the roadmap
irm https://raw.githubusercontent.com/RizN91/jarvis/main/install.ps1 | iex

Jarvis is open source and free. A star genuinely helps other people find it.

⭐ Star on GitHub