Rina Assistant

Say it, and I'll do it.

I'm Rina, a voice assistant for Windows. I open programs, set timers and reminders, and answer questions. I recognize your voice right on your computer.

LunaEverything she remembers about you stays with you. You can open any entry and delete it — she has a page just for that.

Tell me

Windows 10 and 11 · no account · free beta · Apache-2.0

“This is how I hear you: from silence to a line in the journal.”

What happens after you speak

Five steps, and at any of them I can stop and ask. None of them leaves your computer.

  1. Heard

    I cut the phrase out of the silence and turn it into text. Vosk and Whisper work offline.

  2. Understood

    I work out what to do and with what: which program to open, how long to set the timer for. An unfamiliar phrase I can hand to a local model or a web search, but both stay off until you turn them on.

  3. Checked

    Several similar programs, an unsigned file, something that can't be undone — I'll ask and tell you what will happen.

  4. Acted

    Every action goes through a single registry in my core, and that's where each tool's permissions are checked.

  5. Logged

    The journal gets the call: which tool, with what, and how it ended. Free text is kept only as its length.

What I can do

I work by voice from the first launch, nothing to set up. Unless you want to :)

“I find programs on my own. If I'm not sure, I ask.”

I find programs on my own

I look through Start menu shortcuts, the registry, Store apps and any folders you show me. Say “open telegram” and I'll find Telegram, and if I have to ask which one you meant, I'll remember your answer.

Timers, reminders and the system

Timers and reminders go off even when my window is minimized. Volume, pause, screenshot, sleep, shutdown, other programs' windows — “close Discord”, “minimize all windows”. Nineteen system actions in all.

Voice both ways

I answer sentence by sentence without waiting for the end, and you can cut me off with my name or “stop”. I listen with Vosk or Whisper and speak with Edge, gTTS, Piper or pyttsx3 voices.

Your own commands

A command is built from steps: pause, repeat, condition, calling another command. One brought over from another machine reaches me switched off, and you can see all its steps before turning it on.

Plugins

Every plugin runs in its own process, and I check its permissions the same way as for my own actions. Two come with me: Weather and Exchange rates. How to write your own is in the reference RU.

Close at hand

A floating command bar over your windows, the tray, autostart and hotkeys. Three window finishes and two languages: English and Russian.

“Your voice stays on this computer. The rest leaves only if you ask.”

Nothing leaves until you ask

LunaI'm not interested in what you talk to Rina about. There's no account in the beta, but there is telemetry, off by default. In the Stable release it will vanish.

RinaAnd if you'd like to check, it's all visible inside me, on the “What Rina knows about me” page.

Always stays with you

  • Recorded sound and recognized text: Vosk and Whisper run on your computer.
  • The contents of your files and notes, and the conversation history.
  • Paths on your disk, program names, tokens and passwords.
  • The diagnostic package: it is put together only when you ask, and anonymized.

Leaves only if you turn it on

Edge or gTTS voice
the reply text — to Microsoft or Google
Web search for an unfamiliar phrase
the phrase itself — to the search engine in settings: DuckDuckGo, Brave Search, Startpage, Mojeek or Qwant
Local model
the question and a few recent lines — to the address in settings, usually this same computer: Ollama, LM Studio or llama.cpp; an Ollama cloud model forwards them to ollama.com, and I warn you when one is chosen
A model on OpenRouter
the question and a few recent lines — to openrouter.ai, with your key
Web lookup for the local model
the query the model wrote — DuckDuckGo
Model downloads
alphacephei.com, huggingface.co
Update check
api.github.com — on the button or once a day
Weather plugin
the city’s name and its coordinates — open-meteo.com
Exchange rates plugin
cbr.ru
Beta telemetry
numbers once a day, no words, paths or names — rina-telemetry.vercel.app

You turn each of these on yourself, and each one is labeled in the settings.

“What Rina knows about me”

It's a page inside me. It lists learned words, the conversation history, reminders, to-dos, your own commands, plugins, the folders searched for programs, saved sign-ins (that they exist, never the passwords themselves), every telemetry report and the settings you changed. Next to them is a button to make me forget: everything at once or one by one.

Threat model RU

“The installer is out, and I also run from source.”

Download

I install into your profile without administrator rights — I don't need them.

The installer

Beta 4.0.0 is out.

Download the installer

By installing me you accept the licence agreement. What I keep and what I send is in the privacy policy.

Windows, 64-bit, 125 MB. The description and all files are on the release page.

SHA-256: f83ba5c3d9b1b2e9ba7ec58141cebbe314ca8ce5108bbe4e9362740c0abbff89.

LunaWindows will warn about an unknown publisher: I haven't signed the build yet, signing comes with Stable. Click “More info” → “Run anyway”.

RinaI don't touch PATH, don't install a service and don't add myself to autostart: autostart is a setting inside me.

LunaHey! That was my line.

RinaHehe <3

RinaBy the way, you can think of the installer as a present under the tree. Unwrapping is on you :)

From source, today

You need Windows, Python 3.10 or newer, and .NET 9.

git clone https://github.com/Luna-coreX/RinaAssistant.git
cd RinaAssistant
pip install -r requirements.txt
dotnet run --project shell/Rina.Shell

On first launch I'll ask a couple of questions — which voice and which recognition you want — and download only those. Then say or type “open notepad”, “set a timer for 5 minutes” or “what is 17 times 23”.

System requirements

MinimumRecommended
SystemWindows 10 version 2004 or newer, Windows 11; 64-bit only
Processorx64, 2 coresx64, 4 cores or more
Memory4 GB8 GB
Disk spaceabout 1 GB: the program and the small recognition modelabout 3 GB: Whisper and the Piper voice
Sounda microphone; headphones or speakersa headset, or speakers with echo cancellation on
Networkneeded to download models and for the Edge voice; Vosk, Whisper and Piper work without it

A language model of your own is optional. If you want one, it sets its own requirements: a 7–⁠8B model usually needs 16 GB of memory or a graphics card with 8 GB.

“Free? — Yes! But not all of it.”

What it costs

First things first: my whole local base is free.

The beta is free

Completely. Everything the beta has, and everything it still gets, stays free. The rest will depend on time and people.

The local base stays free

Voice both ways, the local model, finding and launching programs, reminders and timers, hotkeys, system actions with confirmation, history, your own commands, built-in plugins and yours, the security journal, the privacy page.

Pro — in 4.0.0 Stable, together with accounts

Sync between devices with end-to-end encryption, ready-made sets of scenarios, connections to third-party services, advanced features... An account is needed only for what's paid; without one, everything works as before.

Cloud — separately, for services

Cloud services through us, such as models via OpenRouter. Connecting OpenRouter with your own key, as in the beta, stays possible — free and without an account.

What stays open

The code that handles sound and your data, launching programs, confirming dangerous actions, the journals, the protocol and the basic features. The Apache-2.0 license applies to the copy you already have, and it can't be taken back after the fact.

LunaHonestly, up front: the paid parts, the servers and billing will be closed, but there will be no closed code inside the open part.

RinaSo you can just take me and use me :)

LunaYou make it sound like you're a thing...

RinaAren't I?

“Next I get a voice of my own and a long memory.”

What's next

The whole plan is public: milestones, blocks and every task, with its size, dependencies and status.

  1. 4.0-port · done

    Separation

    I split into two programs: the core in Python, the shell in C#, with a protocol between them.

  2. 4.0-beta · now

    Open beta

    The installer, this site, privacy and explaining what I do. Right now we're finding out whether I'm useful in real work.

  3. 4.0 Stable

    Release

    Update channels and signing, our own update server, accounts and Pro. The beta telemetry goes away.

  4. 4.1+

    Later

    Rolling back versions and delivery through a CDN.

  5. 5.0

    Platform

    I'll start talking while the answer is still forming. Long-term memory arrives. Computer control — only under detailed permissions.

  6. Track V · in parallel

    Voice

    A voice of my very own.

LunaNext up is Stable: signed releases, updates and accounts for those who want Pro.

RinaAnd I overheard that in 5.0 I'll learn to look at the screen. Eyes! I can't wait.

LunaOverheard, huh... Well, yes, it's true.

Rina “This is what I look like inside... Lights and all, it’s the holidays!”

You'll say the rest.

So, shall we try? Beta 4.0.0 is already out...

LunaIf something breaks, write to me. I read every message.

Rina(If he wakes up, that is. Workaholic.)

RinaAnd happy New Year! I can't decorate a tree, but I can remind you about the presents. Just ask.

LunaHappy New Year from me too!