What Is Voxa?
Voxa is an open-source desktop voice assistant, published at askvoxa.dev and built with Tauri v2, a framework for building lightweight cross-platform desktop apps with a web-based UI wrapped in a native shell. Voxa is released under the MIT license.
- Core interaction model: a frameless, always-on-top orb that floats above your other windows — you tap it to start a realtime voice conversation, rather than opening a separate chat window or browser tab
- Model backend is configurable, not fixed to one provider: cloud realtime voice models (Google Gemini Live or OpenAI Realtime) or a self-hosted local daemon
- Built with Tauri v2, which renders the UI using the operating system's own web engine (WebView2 on Windows, WKWebView on macOS, WebKitGTK 4.1 on Linux) instead of bundling a separate browser engine like Electron does
- Notes and information the assistant looks up or is asked to save are written as plain Markdown files on the local filesystem
- No proprietary backend server run by Voxa itself — the app's own configuration, data, and API keys stay on your machine; any cloud calls go directly to the model provider you configure (Google or OpenAI), not to a Voxa-operated server
What Are Voxa's Core Features?
Voxa's three defining features are the always-on-top voice orb, Markdown-based note storage, and its open-source Tauri v2 foundation. Each shapes how the app behaves compared with a browser-based or subscription voice assistant.
The floating voice orb
Voxa's interface is a frameless, always-on-top orb rather than a full application window or browser tab. It stays visible above other windows so a realtime voice conversation can start with a single tap, without switching focus away from whatever you were working on.
Markdown-based notes and search
When Voxa searches for information or is asked to save something, the result is written as a plain Markdown file on your local filesystem — not locked into a proprietary database or a cloud-only note store. Because Markdown is a plain-text format, those files stay readable and editable in any text editor or note app, independent of whether Voxa is running.
Open source: Tauri v2 and MIT license
Voxa is built with Tauri v2 and released under the MIT license, a permissive open-source license with no restriction on inspecting, modifying, or redistributing the source. Tauri apps use the operating system's own web-rendering engine for the UI instead of bundling a separate browser engine, which is why supported-platform requirements are tied to a specific OS web-engine version on each platform (see the platform section below).
Is Voxa Actually Local?
Voxa is a hybrid app, not an exclusively on-device one — be clear about this distinction before choosing it as a "local AI" tool. It supports two separate backend paths for the realtime voice model, and only one of them keeps model inference on your own hardware.
- Cloud realtime voice models — Voxa can route the realtime voice conversation through Google Gemini Live or OpenAI Realtime. This path requires an internet connection and your own API access/credentials with that provider; the voice model itself runs on that provider's servers, not on your device.
- Self-hosted local daemon — Voxa can instead be pointed at a local daemon you configure and run yourself, keeping voice-model inference on your own machine or network rather than sending audio to a cloud provider. This is the path that makes Voxa function as a local AI tool, but it is opt-in setup work on your part, not the default, zero-configuration behavior.
📌Note: What stays local regardless of which backend you choose: the app's own configuration, your notes (saved as plain Markdown files), and your API keys — these are stored on your machine, not on a Voxa-operated server. What does not stay local by default: the realtime voice model itself, unless you have configured the self-hosted local-daemon path.
Voxa on Windows, macOS, and Linux
Windows
- What to expect:
- Voxa renders its UI using WebView2, the Chromium-based web engine Microsoft ships with modern Windows. A current WebView2 runtime is required for the app's orb interface and voice UI to display correctly.
macOS
- What to expect:
- Voxa uses WKWebView, Apple's built-in web-rendering engine, for its UI. Because Tauri relies on the OS-provided engine rather than bundling its own, behavior tracks whatever WKWebView version ships with your macOS release.
Linux
- What to expect:
- Voxa uses WebKitGTK 4.1 for its UI on Linux. This is a specific WebKitGTK version, not just "any WebKit build" — confirm your distribution packages WebKitGTK 4.1 (or a compatible version) before installing, since older distro package versions may not match.
Because Tauri v2 renders the UI through each operating system's own web engine instead of a bundled browser engine, the exact minimum OS/engine version for each platform is not something this article can pin down as a permanent number — confirm current platform requirements directly at askvoxa.dev before installing.
Who Should Use Voxa?
The right fit depends on whether an always-visible voice orb and Markdown-based notes match how you want to talk to an assistant, and whether you are willing to set up a local daemon if full local inference matters to you.
Use Voxa if
- You want a voice assistant that stays visible as a small floating orb rather than a full window or browser tab you have to switch to
- You want your notes stored as plain Markdown files you can open, search, and edit outside the app, instead of a proprietary format
- You are comfortable using a cloud realtime voice provider (Gemini Live or OpenAI Realtime) with your own API access, or you are willing to configure a self-hosted local daemon if you want inference to stay on your own hardware
- You want an open-source, MIT-licensed app where you can inspect or modify the source, rather than a closed-source subscription assistant
Consider a different tool if
- You need a fully local, zero-configuration voice assistant out of the box — Voxa's default path is cloud-backed, and local-only operation requires you to set up the self-hosted daemon yourself
- You want a mobile app rather than a desktop tool — Voxa targets Windows, macOS, and Linux desktop, not iOS or Android
- You need a fully documented, end-to-end open-source voice pipeline (speech-to-text, LLM, text-to-speech) with every component specified — see Build a Local Voice Assistant on Your Phone or Build a Local Voice Assistant: Full Guide for that kind of fully specified pipeline
Common Mistakes When Evaluating Voxa
Most confusion about Voxa comes from assuming it behaves like either a purely local app or a purely cloud app, when it is designed as a configurable hybrid of both.
Mistake 1: Assuming Voxa runs fully offline by default
Voxa's default, lowest-friction path is a cloud realtime voice model (Gemini Live or OpenAI Realtime), which needs an internet connection and that provider's API access. Offline, fully local operation is possible, but only after you configure and run the self-hosted local daemon yourself.
Mistake 2: Assuming "open source" means "no cloud calls"
Voxa's own application code being open source (MIT license, Tauri v2) is a separate fact from which voice-model backend you choose to use. Open-source app code can still be configured to send realtime audio to a third-party cloud provider — check which backend path you have actually configured, not just the app's license.
Mistake 3: Expecting the same Markdown notes across every device automatically
Notes are saved as plain Markdown files on the local filesystem of the machine Voxa is running on. Unless you separately set up your own sync (a cloud-drive folder, a git repo, or similar), notes on one machine are not automatically available on another.
Mistake 4: Installing without checking the platform web-engine requirement
Because Voxa is built with Tauri v2, its UI depends on the operating system's own web engine — WebView2 on Windows, WKWebView on macOS, WebKitGTK 4.1 on Linux — rather than a bundled browser engine. An outdated or missing engine on your system can prevent the orb UI from rendering correctly; confirm your platform's engine is current before assuming an install issue is a Voxa bug.
Frequently Asked Questions
What is Voxa?
Voxa is an open-source, MIT-licensed desktop voice assistant built with Tauri v2, available at askvoxa.dev. It presents itself as a frameless, always-on-top orb you tap to start a realtime voice conversation, and it can use a cloud realtime voice model or a self-hosted local daemon as its backend.
Is Voxa free?
Voxa's own application code is open source under the MIT license. If you use the cloud realtime voice backend (Gemini Live or OpenAI Realtime), you are subject to that provider's own pricing and API terms, separate from Voxa itself. Confirm current pricing for Gemini Live or OpenAI Realtime directly with Google or OpenAI.
Does Voxa run fully offline?
Only if you configure the self-hosted local daemon path. By default, Voxa is set up to use a cloud realtime voice provider (Gemini Live or OpenAI Realtime), which requires an internet connection. A fully offline setup is possible but requires you to run and point Voxa at your own local daemon.
What platforms does Voxa support?
Windows (using the WebView2 engine), macOS (using WKWebView), and Linux (using WebKitGTK 4.1). Voxa is a desktop app built with Tauri v2, not a mobile app for iOS or Android.
Where does Voxa store my notes?
As plain Markdown files on your local filesystem, not in a proprietary database or a cloud-only format. This means notes remain readable and editable in any text editor independent of whether Voxa is installed or running.
Does Voxa send my API keys or data anywhere?
Configuration, data, and API keys are stored on your own machine. If you use the cloud realtime voice backend, your audio and conversation data is sent to whichever provider you configured (Google Gemini Live or OpenAI Realtime), under that provider's own data-handling terms — not to a separate Voxa-operated server.
What is the difference between the cloud backend and the local daemon backend in Voxa?
The cloud backend (Gemini Live or OpenAI Realtime) routes the realtime voice model through that provider's servers and requires an internet connection and API access with that provider. The self-hosted local daemon backend keeps voice-model inference on hardware you control, but you have to set up and run that daemon yourself — it is not preconfigured.
Is Voxa open source?
Yes. Voxa is built with Tauri v2 and released under the MIT license, a permissive license that allows inspecting, modifying, and redistributing the source code without restriction on commercial use.
Can I use Voxa without an internet connection?
Only if you have configured and are running the self-hosted local daemon as your voice-model backend. The cloud realtime-voice path (Gemini Live or OpenAI Realtime) requires connectivity to function.
Does Voxa require a specific browser engine on my system?
Yes, indirectly — because it is built with Tauri v2, Voxa renders its UI through the operating system's own web engine: WebView2 on Windows, WKWebView on macOS, and WebKitGTK 4.1 on Linux. Make sure the relevant engine is present and current on your system before installing.