Installation

NoteTalker is a native desktop app for Windows, macOS, and Linux. The AI runs entirely on your own machine, so the app needs a capable GPU and some disk space for the models it downloads on first launch. Installation itself takes just a few minutes.

System Requirements

Operating system

Platform	Minimum Version
Windows	Windows 10 (64-bit) or later
macOS	macOS 12 Monterey or later
Linux	Ubuntu 20.04+, Debian 11+, or any distro supporting AppImage

Hardware for the local AI

NoteTalker runs a language model, speech-to-text, and text-to-speech on your device — nothing is sent to a server — so a supported GPU is required to run the chat model:

• Windows and Linux: an NVIDIA GPU with CUDA.
• macOS: Apple Silicon (M1 or later), which runs the model on MLX/Metal.

How much video memory (VRAM) you have decides which model you can run. The default model is Mistral Nemo 12B; on a smaller GPU, choose the lighter Qwen3 4B in Settings → AI.

Your GPU / memory	Recommended model
NVIDIA, ~10–12 GB VRAM or more	Mistral Nemo 12B (the default)
NVIDIA, 8 GB VRAM	Qwen3 4B
Apple Silicon, 24 GB or more unified memory	Mistral Nemo 12B
Apple Silicon, 16 GB unified memory	Qwen3 4B

The language model (chat) runs on CUDA (Windows/Linux) or MLX/Metal (macOS) — there is no Vulkan backend. The search and voice components (embeddings, wake word detection, speech-to-text) run on ONNX Runtime and can use CUDA, DirectML, or CoreML for acceleration, but fall back to CPU if no compatible GPU is present — so search and voice still work on any machine, just more slowly.

Disk space

The NoteTalker installer itself is small. The AI models are downloaded separately the first time you launch the app:

• ~3 GB for the essential set — the smallest language model (Qwen3 4B), Whisper Base for speech-to-text, and Kokoro for text-to-speech.
• ~8 GB if you use the default Mistral Nemo 12B model instead of the 4B.

Leave room for your vault and its search index on top of that. Models are stored in a per-OS application-data folder, not in your vault:

Platform	Model location
Windows	%APPDATA%\Atlas-Local\models\
macOS	~/Library/Application Support/Atlas-Local/models/
Linux	~/.config/Atlas-Local/models/

Your vault folder can be as large as you need — it’s just plain text files on your drive.

Download NoteTalker

Download the latest version from either of these sources:

• Official website: talkwithyournotes.com/download
• GitHub Releases: github.com/boringbots/atlas-local-releases

Choose the installer for your operating system from the releases page.

---

Installing on Windows

1. Download the .exe installer file.

2. Double-click the downloaded file to run it.

3. If Windows SmartScreen shows a warning (“Windows protected your PC”), click More info and then Run anyway. This warning appears because NoteTalker is a new app from a small team — it’s safe to proceed.

4. Follow the on-screen installer steps. NoteTalker installs to your user folder by default (no admin rights needed).

5. Launch NoteTalker from the Start menu or the desktop shortcut.

Tip

If your organization’s IT policy blocks .exe installers from unknown publishers, ask your IT team to whitelist NoteTalker or check whether a pre-approved distribution channel is available.

---

Installing on macOS

1. Download the .dmg disk image file.

2. Double-click the .dmg to mount it, then drag the NoteTalker icon into your Applications folder.

3. The first time you open NoteTalker, macOS Gatekeeper will block it because it wasn’t downloaded from the App Store.

   To open it anyway: find NoteTalker in your Applications folder, right-click (or Control-click) the app icon, and select Open from the menu. Click Open in the dialog that appears.

4. After this first-time approval, NoteTalker will open normally on every subsequent launch.

Caution

Do not try to open NoteTalker by double-clicking it the first time — Gatekeeper will only give you the option to move it to trash. You must use right-click > Open to get the option to approve it.

---

Installing on Linux

AppImage (recommended, works on most distros)

1. Download the .AppImage file.

2. Make it executable. Open a terminal and run:

   chmod +x NoteTalker_*.AppImage

3. Run it:

   ./NoteTalker_*.AppImage

You can move the AppImage anywhere on your system — your home folder, /opt, or wherever you prefer. No installation needed.

Debian/Ubuntu (.deb package)

1. Download the .deb file.

2. Install it with dpkg:

   sudo dpkg -i atlas_*.deb

3. Launch NoteTalker from your application menu or run atlas in a terminal.

Linux system packages

NoteTalker on Linux uses WebKit2GTK for its interface. If the app fails to launch, make sure libwebkit2gtk-4.1 is installed. For the best text-to-speech pronunciation, also install espeak-ng (recommended) — NoteTalker uses it to phonemize speech. On Debian/Ubuntu:

sudo apt install libwebkit2gtk-4.1-0 espeak-ng

espeak-ng is optional; without it, voice output still works but pronunciation of some words is rougher. You can also install it later from the guided setup in Settings → Voice.

GPU drivers on Linux

The local language model requires an NVIDIA GPU with the CUDA toolkit installed. To run it, you’ll need:

• NVIDIA GPU (Maxwell architecture or newer)
• NVIDIA driver 525 or later
• CUDA Toolkit 12.x (libcublas, libcurand, libcufft, libcudart)
• cuDNN 9.x (libcudnn)

The search and voice components can also use CUDA when it’s present, and fall back to CPU when it isn’t — so those keep working even without the toolkit, just more slowly. The chat model, however, needs the GPU.

---

First Launch

Setting up NoteTalker the first time is a quick, three-step flow. Only the license activation and the model download need the internet — after that, NoteTalker runs fully offline.

1. Choose a vault folder. This is where all your notes and data are stored.

   • Create a new folder if you’re starting fresh (e.g., ~/My NoteTalker Vault).
   • Choose an existing folder if you already have markdown notes you want to use.

   NoteTalker sets up its own atlas/ subfolder inside whatever folder you choose. Your existing files won’t be touched. See Your First Vault for what NoteTalker creates and how your vault is organized.

2. Activate your license. Paste the license key from your account into Settings → License & About. Activation is a one-time online step that binds the license to this computer; after that, NoteTalker works offline forever. Your $37 purchase covers all 2.x releases and up to two devices. See License & Activation for buying, moving to a new computer, and troubleshooting activation.

3. Download the AI models. NoteTalker downloads the models it needs (the essential set is ~3 GB; the default Mistral Nemo 12B brings it to ~8 GB) and shows a progress bar as it goes. The downloads are resumable, so a dropped connection picks up where it left off — on a typical home broadband connection this takes anywhere from a few minutes to about a quarter of an hour, depending on the model you choose. You can start using your notes right away; the AI features become available once the download finishes.

---

Updating NoteTalker

NoteTalker does not download or install new versions on its own — each update is a manual download. When a new 2.x release is available, grab the latest installer for your platform from the download page or GitHub Releases and install it over your current version. Your vault and your license activation are untouched.

Tip

Your license covers every 2.x release, so updates within the 2.x line are free — there’s nothing to re-purchase and nothing to re-activate.