Private beta Local LLM · Free · Works offline
Free local AI for your computer.
Twoody runs a real language model on your computer, Mac, Windows or Linux: your own AI, at home, for you as for your company. Chat, write, summarize and ask questions about your documents: no AI in the cloud, it works with Wi-Fi off, and there is no ChatGPT, Claude or Gemini subscription to pay.
- Mac
- Windows
- Linux
Private beta · Mac (macOS 13 or later), Windows 10 and 11, Linux: free, no account · iPhone and Android, with your computer. Which model for my computer?

- The AI runs 100% on your computer
- Works offline
- Free, no account
- Notarized on Mac, signed on Windows
Why local AI
Your AI, on your own computer.
An online assistant runs on someone else's servers: your questions and your files go there, it needs a connection, and it costs every month. A local AI runs on the machine in front of you.
Private by design
The model runs on your computer: no AI service receives your conversations or your documents. Without an account, or in a private conversation, nothing leaves it at all.
Works offline
Once the model is installed, turn Wi-Fi off: Twoody keeps answering. On a plane, on a train, or on a network that must stay closed.
Free, with no quota
No subscription, no account needed, no message limit. The only limit is the speed of your computer.
Open-weight models
Qwen3.5, published under the Apache 2.0 licence and converted by Osmove for Twoody's engine, installed in one click and sized for your computer's memory. Or the models you already run in LM Studio or Ollama, and in mlx-lm on a Mac.
Replace a subscription
Everyday AI, without a subscription.
For writing, summarizing and working on your own files, a model on your computer does the job. Here is what changes compared with the big online assistants.
| Replace a subscription | Twoody | ChatGPT Plus | Claude Pro | Google AI Pro |
|---|---|---|---|---|
| Price | Free | Paid monthly subscription | Paid monthly subscription | Paid monthly subscription |
| Where your conversations go | On your computer, or on your account if you choose | To OpenAI's servers | To Anthropic's servers | To Google's servers |
| Works without internet | Yes, once the model is installed | No | No | No |
| Account needed | No | Yes | Yes | Yes |
| Usage limits | None but your computer's speed | Set by OpenAI | Set by Anthropic | Set by Google |
| The model | Qwen3.5, 4B or 9B, on your computer | Large models, in data centers | Large models, in data centers | Large models, in data centers |
The large online models remain stronger for the hardest questions: Twoody is made for everyday work, on your own files, in private.
What Twoody does
What you ask an AI, on your computer.
Chat and write
Ask anything, draft an email, summarize, rephrase, translate. For a hard question, choose a greater effort: Thinking or Deep.
Your documents, with their sources
Attach a file or choose folders, down to your iCloud Drive, Google Drive, Dropbox or OneDrive one: Twoody reads PDF (scanned ones included, on the Mac), Word, OpenDocument, RTF, HTML, text, Markdown and CSV files on your computer, looks at them only when a question needs them, and its answers cite their sources, numbered.
Memories
What Twoody should remember about you and your work, from one conversation to the next: what you ask it to remember, and what it learns by reading your conversations again — what it is unsure of, it suggests first (“To confirm”). Give each memory its importance, import what ChatGPT, Claude, Gemini or Copilot know about you, see and erase each memory.
Projects and reminders
Twoody notices the projects you talk about — a purchase, a trip, a case — and offers to follow them, with their size and deadline. “Remind me on Friday to call the notary” creates a reminder, which rings on your computer, or a task.
Mac, Windows and Linux
The same app on all three systems. The Mac also reads scanned PDFs; on Windows, the model uses the graphics card through Vulkan when its driver allows; on Linux, it runs on the processor. The iPhone and Android apps use your computer's model. All of it is in private beta.
Several computers (beta)
A Raspberry Pi 5 or a headless server joins your account with a code: the desktop app can welcome it already, and installing Twoody on such a server will be published soon. “Automatic” sends each question to the computer that will answer fastest, and another computer receives your memories and looks at your documents only if you trust it.
Agent (beta)
Hand over a task on a folder: find across every document, compare, sort the files, draft a Word document. It asks before every change and deletes nothing.
Twoody Code (beta)
A coding agent that works in a folder, with a terminal for the servers it keeps running, and which you can also drive from the iPhone if you turn it on. By default, it asks before it changes a file or runs a command. A local model stays far behind the large online models at code: hand it well-bounded tasks.
Voice messages
Speak instead of typing: your voice is transcribed on the computer, by macOS speech recognition on the Mac, and by whisper.cpp, shipped with the app, on Windows and Linux or when needed. On the iPhone, a voice message is transcribed on the phone, and only its words go to your computer.
Contacts and Calendar
On the Mac only, if you turn them on: the chat finds the card of a person you name, or your appointments. They are read on the Mac, never changed.
For the household and the company
Bring the people you live or work with into a household or an organization, and share a computer's model with them if you choose to. Being part of it gives them nothing of your data.
Your usage, at home
Tokens per day, week, month or year, the speed of your models and the computer's load; under each answer, ⋮ shows what the model received, where the time went and the share served from the cache. Counted on the computer; only the diagnostics' measurements without content leave it, during the beta, if you leave them on.
Your models too
Already using LM Studio or Ollama, or mlx-lm on a Mac? Twoody uses the models they run, next to its own.
Web search, if you want it
Off at first. Turned on, it sends your search, never your documents, to the engine you choose.
Connectors (beta)
Plug in your services through MCP servers — picked from a list (Notion, GitHub, Linear, Stripe…) or added by hand — and the chat uses their tools. You choose the tools it sees, and it asks you before each call, except for those you allow once and for all. Off at first.
Home (Home Assistant)
Link your Home Assistant, and say in the chat “turn on the living room light”: Twoody controls your devices from your computer. Off at first.
Not yet: Excel and PowerPoint files, and installing Twoody on a headless server. A very faint or blurred scan can escape recognition.

How it works
Three steps, a few minutes.
-
Download
Pick the version for your system: Mac (Apple Silicon or Intel, notarized by Apple), Windows (an installer signed by Osmove) or Linux (AppImage or .deb package). Install it like any other app.
-
Install a model
Twoody offers the model that suits your computer's memory. One click, and it downloads: 2.8 or 5.8 GB.
-
Ask
The conversation opens. Ask your first question, attach a document, and turn Wi-Fi off if you like.
Requirements
Which model for my computer?
The larger the model, the finer its answers, and the more memory it needs. Twoody offers the right one for your computer.
- Mac: macOS 13 (Ventura) or later.
- An Apple Silicon Mac (M1 and later) is recommended. Intel Macs work, more slowly: the model runs on the processor.
- Windows 10 or 11, 64-bit (Intel or AMD): the model runs on the graphics card through Vulkan when its driver allows, otherwise on the processor.
- Linux x86_64 (AppImage or .deb package, tested on Ubuntu 24.04): the model runs on the processor.
- Disk space for the model: 2.8 or 5.8 GB.
| Memory | Model | Download |
|---|---|---|
| 8 GB and more | Qwen3.5 4B | 2.8 GB |
| 16 GB and more | Qwen3.5 9B | 5.8 GB |
Private beta
Twoody is in private beta
Twoody runs on Mac, Windows and Linux, and its iPhone and Android apps work with Twoody on your computer; a first circle of users is testing it all before it opens to everyone. Join the waitlist: we will write to you when Twoody opens to you. It stays free, and on the computer it needs no account.
- Mac: macOS 13 or later, Apple Silicon or Intel.
- Windows 10 and 11, 64-bit (Intel or AMD).
- Linux x86_64: AppImage or .deb package (Debian, Ubuntu).
- iPhone and Android: the app works with Twoody on your computer, and uses its model.
On all your devices
Your computer, and your phone with it.
Twoody is in private beta on Mac, Windows and Linux, and on iPhone and Android: the phone apps use the model running on your computer and let you pick up your conversations.
- Mac Private beta
- Windows and Linux Private beta
- iPhone and Android Private beta
Join the waitlist
For professionals
Bound by confidentiality? Your files stay in the office.
Lawyers, doctors, accountants, consultants: with an online assistant, every document you give it goes through another company's servers. With Twoody, it is read on your computer by a model installed there, and in a private conversation nothing about it leaves the computer.
- Lawyers and in-house counsel
- Doctors and health professionals
- Accountants and advisers
- Consultants and executives
Privacy
What leaves your computer, and nothing else.
The AI runs 100% on your computer. Without an account, your conversations, your memories and the index of your documents stay there; with the optional account, the conversations you keep on it go to it, encrypted end to end. During the beta, technical diagnostics without content, which can be turned off; no crash reports, no ad tracking. Here is everything that goes out.
-
The update check
At most once a day, the app reads a small file on downloads.twoody.com. You can turn it off in Settings.
-
The model, when you install it
It is downloaded from Hugging Face, then checked before it is used.
-
The voice model, the first time you speak
The model of whisper.cpp (574 MB on an Apple Silicon Mac) is downloaded from Hugging Face on the first press of the microphone.
-
Your web searches, if you turn them on
The search goes to the engine you chose. Your documents, never.
-
Your Twoody account, if you sign in
Optional. The conversations you keep on your account are stored on twoody.com; private conversations stay on the computer.
-
The encrypted relay, if you open it
Away from home, your devices reach your computer through your account's relay, in exchanges sealed end to end: it sees that they talk, when and how much, never what they say.
-
Diagnostics without content, during the beta
On by default during the beta, can be turned off in Settings: every 15 minutes, measurements of the answers (model, durations, token counts), never a question, an answer or a document. With an account they go to the account; without one, to twoody.com under a number drawn at random by the installation.
-
The whole of each answer, if you tick it
Unticked by default. Ticked, each answer goes whole — your question, the instructions given to the model, the tools, the answer —, to help us audit and improve Twoody.
-
The connectors you add
Off by default. A tool receives only the arguments shown, at each call you allow, and only the service you added.
-
Your Home Assistant, if you turn it on
The commands you ask for go from your computer to your Home Assistant, at the address you give.
Frequently asked questions
Is Twoody really free? How is it funded?
Yes, for personal and professional use. Twoody is published and funded by Osmove, a French artificial intelligence consultancy. No advertising, no data sold, and no paid offer.
Can Twoody replace ChatGPT, Claude or Gemini?
For everyday work — writing, summarizing, translating, questions about your own documents — often yes, for free and in private. A model that runs on a personal computer is smaller than the large online models: for the hardest questions, they remain stronger.
Does it work without an internet connection?
Yes, once the model is installed: turn Wi-Fi off and Twoody keeps answering and reading your documents. Only web search and the conversations you keep on your optional account need the internet.
What is a local LLM?
A large language model (LLM) is the kind of AI behind ChatGPT. A local LLM runs on your own computer instead of a company's servers: no AI company receives your questions, and it works offline. Twoody installs one for you, in one click.
How much memory does my computer need?
8 GB is enough for Qwen3.5 4B; from 16 GB, Twoody installs Qwen3.5 9B, which gives finer answers. With more memory, you can use larger models through LM Studio or Ollama.
Does it work on an Intel Mac, or on macOS 13?
Yes. Twoody runs on macOS 13 (Ventura) or later, on Apple Silicon and on Intel. On an Intel Mac the model runs on the processor: it works, more slowly.
Which systems does Twoody run on? And on iPhone or Android?
The same app runs on Mac (macOS 13 or later), Windows (10 and 11, 64-bit) and Linux (x86_64). The Mac remains the most complete version: only it reads scanned PDFs, Contacts and Calendar, and dictates with macOS speech recognition. On iPhone and Android, the app uses the model of your computer. All of it is in private beta, not yet a public download: leave your email to join the waitlist.
Can I find my conversations on my other computers, or on my phone?
On your other computers, Mac, Windows or Linux, yes, with the optional Twoody account, which is free: the conversations you keep on it open on the computers signed in to the same account, and private conversations stay on their computer. On a phone, the iPhone and Android apps, in private beta, work with Twoody on your computer and let you pick up your conversations.
Can several computers work together, such as a Raspberry Pi?
Yes, with the optional Twoody account: a computer running Twoody without a screen, such as a Raspberry Pi 5, joins your account with a code, and “Automatic” sends each question to the one that will answer fastest. A model is never split across machines: each question goes to a single computer. By default, another computer receives only the conversation, without your memories or your documents, and what your computers exchange is encrypted end to end. The desktop app can already welcome it; installing Twoody on a headless server will be published soon.
Are my documents sent anywhere?
Without an account, or in a private conversation, no: they are read on your computer, by a model installed on it, and nothing about them leaves it. In a conversation you keep on your optional Twoody account, your messages, the answers and the passages Twoody read are kept on the account, encrypted end to end on your devices, so that you find them on your other devices: our servers cannot read them. Web search, off by default, sends only your search; a connector you add receives only what you allow it to receive.
Where does the model come from?
Qwen3.5 is published by Alibaba's Qwen team, under the Apache 2.0 licence. Osmove converts it for llama.cpp, the engine inside Twoody, and publishes the files on Hugging Face; Twoody downloads the one that suits your computer and checks its fingerprint before using it. Once installed, it is a file on your computer, and the engine only listens to your computer.
Which languages does it speak?
Twoody's interface is in English and in French. The model understands and answers in many more: Qwen3.5 supports 201 languages and dialects.
Can I use it at work, for my clients?
Yes: Twoody is free for personal and professional use. Your clients' documents are read on your computer and go to no AI service; keep their conversations private, and nothing about them leaves the computer.
Can I rely on the answers?
A language model can be wrong, even when it cites its sources, and a model that runs on a personal computer is smaller than the large online models. Check its answers, especially before a professional decision.
Are my conversations backed up?
Twoody keeps them on your computer, in its data folder, and makes no backup of its own: if you back up your computer, the backup includes them. The conversations you keep on your optional account are there too, encrypted end to end. Encrypt your computer's disk: FileVault on a Mac, BitLocker on Windows, LUKS on Linux.
Is Twoody open source?
No. The app is free, but its code is not published. Models keep their own licence (Apache 2.0 for Qwen3.5) and the llama.cpp engine is under the MIT licence: Settings › About names them, and their texts ship inside the app.
How do I uninstall Twoody and erase everything?
Follow the two steps on this page(in English): the app, then its data folder.
How do I reach you?
From the app, in Settings › About, or at support@twoody.com. Please do not send your clients' documents by email.
Twoody is in private beta.
On Mac, Windows and Linux, free, with no account and no subscription — and on iPhone and Android, with Twoody on your computer. Leave your email: we will write to you when Twoody opens to you.