LifeOS
Local-first companion
Voice-first, privacy-first AI companion for Windows - reminders, chat, web answers, and email, on-device by default.
This project
v0.1.0 · Windows · Free & open-source
Scroll to explore
The AI assistant that stays on your device.
LifeOS is a voice-first companion for Windows named Yogi. Talk or type to set reminders, get answers from the web, and hear your inbox - with no account, no telemetry, and your data living in a file on your own machine.
On-device by default
No account, no sync
No telemetry
Bring-your-own AI key
Free & MIT
How LifeOS compares
One desktop companion vs. a stack of single-purpose tools.
Great tools already exist for each slice of this - a launcher here, a dictation box there, an inbox agent somewhere else. LifeOS folds those workflows into one local-first Windows app. Here is an honest side-by-side, drawn only from each product's official site.
Raycast
Launcher / command bar
An extendable keyboard-driven launcher with built-in AI chat and 1000+ extensions. macOS-first, Windows in beta.
raycast.com
Wispr Flow
Voice dictation
AI voice-to-text that turns speech into polished writing inside the apps you already use, across desktop and mobile.
wisprflow.ai
Superwhisper
Voice dictation
AI voice-to-text with on-device transcription models and screen-aware modes. Works offline on supported hardware.
superwhisper.com
DeskWand
Desktop automation
An open-source, local-first AI desktop agent that automates repeatable computer tasks with your own model and key.
deskwand.com
AskDesk
Windows launcher
A keyboard-first AI productivity launcher for Windows, with AI assist, quick actions, and voice commands on its Pro tier.
getaskdesk.com
Jace
AI email assistant
A cloud AI executive assistant that organizes your inbox, drafts replies, and automates email workflows.
jace.ai| Product | Voice | NL reminders | Desktop launcher | Local-first | Open source | Platforms | Bring your own key | Offline | Pricing | |
|---|---|---|---|---|---|---|---|---|---|---|
| LifeOS |
✓ | ✓ | ✓ | ✓ | ✓ | ✓ | Windows | ✓ | ✓ | Free · MIT |
| Raycast |
✗ | ✗ | ✓ | ✗ | ✗ | ✗ | macOS · Win (beta) · iOS | Partial | Partial | Free / from $8·mo |
| Wispr Flow |
✓ | ✗ | ✗ | ✗ | ✗ | ✗ | Win · macOS · iOS · Android | ✗ | ✗ | Free / from $12·mo |
| Superwhisper |
✓ | ✗ | Partial | ✗ | ✓ | ✗ | Win · macOS · iOS | ✓ | ✓ | Free / from $8.49·mo |
| DeskWand |
✗ | ✗ | ✗ | ✗ | ✓ | ✓ | Win · macOS · Linux | ✓ | Partial | Free |
| AskDesk |
✓ | ✓ | ✓ | ✗ | ✓ | ✗ | Windows | ✗ | Partial | Free / from $8·mo |
| Jace |
✗ | ✗ | ✗ | ✓ | ✗ | ✗ | Web | ✗ | ✗ | From $20·mo |
✓ Supported
✗ Not offered / not documented
Partial Limited, plan-gated, or beta
Compiled only from each product's official website (researched July 2026). “Not documented” means the official site does not list the feature - it may still exist. Products evolve; verify current details on their sites.
In the box
Everything Yogi can do today.
A focused feature set that actually ships in v0.1.0 - no vaporware.
Voice conversations
Hold a real, persistent conversation with Yogi across auto-titled, reopenable chat sessions.
Natural reminders
Create reminders by talking or typing - Yogi asks instead of guessing when the time is unclear.
Web search with sources
Ask live-fact questions and get answers with citations - and an honest note when a search fails.
Natural voice output
Six voice personalities with a speed control and preview - streamed from OpenAI or offline Windows voices.
On-device speech-to-text
Speech is transcribed by a model that ships inside the app - offline by default, cloud only if you opt in.
Schedules & history
See every active schedule, pause them all at once, and review a filterable history of what fired.
Email summaries
Turn each new email into a chat with a summary, action items, and a spoken heads-up.
Reminders that act
A reminder can run a web lookup when it fires and speak the answer - not just the note you left.
Works offline
Time, greetings, reminders, schedules, and settings all work with zero network and no key.
Under the hood
Complete application flow
Every stage from the download link to a spoken answer - the real path through LifeOS, with the on-device work in the middle and OpenAI touched only when you've turned a cloud feature on.
Get it & launch
Per-user install, no admin.
Download the installer
A single ~155 MB .exe (a portable build is available too).
Install it (per-user, no UAC)
Adds Desktop, Start-Menu & uninstall entries - no admin prompt. (The portable build just runs, no install.)
Launch & single-instance lock
A second launch just focuses the running window - never a duplicate.
Initialize & secure
All local, before any UI shows.
Open & migrate the local database
SQLite via node:sqlite (WAL) opens lifeos.db, runs migrations to schema v8, and seeds 50 settings.
Install session security
A default-deny network filter + CSP go up first. Nothing can reach the internet yet.
Initialize services
Conversation engine, action dispatcher, wall-clock scheduler, provider registry, and Gmail sync are wired up.
Desktop ready
Hidden audio, popup & launcher windows spin up; the tray icon appears and Alt+Shift+Space is armed. Overdue reminders are caught up.
Listen & understand
On-device by default.
You speak or type
Press the mic or the global hotkey. The mic engages only on your action - no wake word, no always-on listening.
Capture → 16 kHz PCM
An AudioWorklet down-samples the mic to 16 kHz and streams frames to the main process. Audio is never written to disk.
Speech-to-text
Transcribed on-device by the bundled sherpa-onnx Zipformer - or, if you opt in, by OpenAI gpt-4o-transcribe with automatic fallback to sherpa.
Intent routing
A local router answers time / greeting / settings instantly with no model. Otherwise the gated LLM (gpt-4o-mini, strict JSON) classifies the turn - and if there's no key, an offline parser takes over.
Reminder
Re-routed to the local parser → a confirmation card. The LLM never creates it; nothing saves until you confirm by click or “yes.”
Chat / question / research
Answered by the model; live-fact turns trigger a web search (gpt-4o-mini-search-preview) and return an answer with sources.
Email tools
When Gmail is connected, a new email is summarized into its own chat with action items - grounded, local, opt-in.
Act & respond
OpenAI only when enabled.
OpenAI - gated
Requests reach api.openai.com only when a capability is enabled, keyed & consented, billed to your account. A reliability guard rewrites any fabricated success into an honest failure.
Voice output
Spoken through a hidden audio window - offline Windows voices by default, or streamed OpenAI gpt-4o-mini-tts via Media Source Extensions when you've enabled cloud voices.
Displayed everywhere at once
The reply fans out to every window - main chat, popup, and launcher stay in sync on one live conversation.
Persist
Locally, on your disk.
Saved to your conversation history
The turn is written to chat_turns, and any fired reminder to history - all in your local SQLite file, never uploaded.
A closer look
One assistant, every surface.
Uniformly framed so nothing is stretched or cropped. Click any shot to zoom.
Why local-first matters
Privacy enforced in code, not in a policy.
Most assistants send everything to a server and ask you to trust a document. LifeOS flips that: it runs on your machine, and cloud access is opt-in, per-feature, and uses your own key.
Your data, one file
Reminders, chats, and settings sit in a local SQLite database you can open and read yourself.
Default-deny network
A built-in filter blocks every outbound request until you switch on a cloud feature.
No tracking, ever
No analytics SDK, no crash reporter, no usage stats, no update ping. LifeOS doesn’t know you exist.
Bring your own key
Cloud features use your OpenAI key, encrypted with Windows secure storage and billed to you.
What cloud AI costs
Free to run. Pennies when you opt in.
LifeOS itself is free. Cloud features use your own OpenAI key and are billed to you directly. Here's roughly what that adds up to - so there are no surprises.
| Capability | OpenAI model | Approx. price | Free offline alternative |
|---|---|---|---|
| Speech-to-text | gpt-4o-transcribe | ~$0.006 / minute of audio | ✓ sherpa-onnx, on-device |
| Chat & responses | gpt-4o-mini | ~$0.15 / 1M input · $0.60 / 1M output tokens | ✗ needs a key |
| Text-to-speech | gpt-4o-mini-tts | ~$0.015 / minute of audio | ✓ Windows voices |
| Web search | gpt-4o-mini-search-preview | gpt-4o-mini tokens + a small per-call search fee | ✗ needs a key |
Our worked estimate assumes a fully-cloud voice exchange: ~1 minute of audio in and out,
one
gpt-4o-mini turn (~1,200 input + ~150 output tokens). STT ≈ $0.006 + chat ≈ $0.0003 +
TTS ≈ $0.0075. Text-only chat, or voice with the offline STT/TTS defaults, costs far less because it skips
the per-minute audio charges.
~$0.014a 1-minute voice exchange
(all cloud features on)
(all cloud features on)
~$0.14a 10-minute conversation
(≈ ten exchanges)
(≈ ten exchanges)
~$1.40100 one-minute conversations
(≈ a light month)
(≈ a light month)
$0.00the same, fully offline
(local STT + Windows voice)
(local STT + Windows voice)
Free forever - no key, no network
- Natural-language reminders, recurrence, snooze & history
- On-device speech-to-text (sherpa-onnx)
- Offline Windows voices
- Local commands - time, date, greetings, help, “show schedules”
- Notifications, schedules, settings, theme, reset
Needs your OpenAI key - opt-in, per feature
- Smarter conversational chat (
gpt-4o-mini) - Higher-accuracy cloud speech-to-text
- Natural cloud voices (streamed)
- Web search with real citations
- Gmail email summaries & auto-research
Pricing changes. These figures are estimates checked against OpenAI's published rates in
July 2026 and are shown only to set expectations. Always verify the current numbers on
OpenAI's official pricing page
before relying on them - your actual cost depends on your usage, prompt length, and OpenAI's current rates.
Download
Get LifeOS for Windows.
Free, open-source, and yours to keep. Install it and start talking to Yogi in minutes.
Heads-up: this is the Windows installer - it sets LifeOS up for
your user only, with no admin prompt. It isn’t code-signed yet, so Windows SmartScreen may show
“Windows protected your PC.” Choose More info → Run anyway to continue. Prefer to build it
yourself? Grab the source on GitHub.
Operating systemWindows 10 / 11
Architecture64-bit (x64)
Disk space~400 MB
Memory~200 MB idle
AccountNone required
PriceFree · MIT
FAQ
Honest answers.
Do my reminders fire when LifeOS is closed?
Reminders fire while LifeOS is running, including when it’s tucked away in the system tray (closing the window with ✕ keeps it running by default). If you fully quit LifeOS or shut down your PC, reminders won’t fire until you open it again - and when you do, it tells you what you missed rather than firing old alarms late. This is deliberate: LifeOS refuses to install a hidden background service.
Is my data sent anywhere?
Not by default - LifeOS makes zero network requests with its default settings, enforced in code by a filter that blocks anything not on an allowlist. The only exception is when you enable a cloud AI feature and add your own OpenAI key; then the text of that command (plus the current date and time) goes directly to OpenAI under your account. Your reminders, history, settings, and audio are never sent.
Do I need an OpenAI key?
No. LifeOS works fully offline for reminders, time, schedules, on-device speech-to-text, and offline voices. A key is optional and unlocks smarter chat, cloud speech-to-text, more natural voices, web search, and email summaries. You bring your own key and OpenAI bills you directly - the app itself is free.
Does Yogi listen in the background or use a wake word?
No. There’s no wake word and no always-on listening. Yogi only records when you press the mic or the global hotkey, and audio is transcribed on your device and never written to disk. A wake word is on the roadmap but is not in this release.
Which platforms are supported?
Windows 10 (version 1809 or newer) and Windows 11, 64-bit only. macOS, Linux, mobile, and web are not available in this release. The core is written in portable TypeScript, so other platforms are a future possibility.
Why does Windows warn me during install?
The installer isn’t code-signed yet, so Microsoft SmartScreen shows a generic “Windows protected your PC” warning. Click More info → Run anyway to proceed. Code signing is planned for a future release.
How do I remove my data?
Open Settings → Danger Zone → “Reset local data.” It permanently deletes the reminders, history, and settings in your LifeOS data folder and nothing else on your device. Note that uninstalling the app does not delete this folder - that’s intentional, so an accidental uninstall never loses your data.