A world without typing
XAIVX AI VOICE
Just speak, and it types
Talk to AI the way you talk to people
Hold the key, speak, let go.
That is all — the text lands in whatever app you have open.
Three steps, that's it
Press and hold the right Alt key
It records only while you hold the key
The text appears in about a second
What you say never leaves your PC
Your recording is never sent over the internet. It is turned into text inside your PC and discarded the moment that finishes. Nothing is stored on our servers. That is why you can dictate confidential material and customer names without worrying.
Note: your audio is never transmitted, whatever your settings. Only if you turn on AI cleanup — which is off by default — is the transcribed text sent to the AI provider you chose.
Vibe-code with your voice
Claude Code, Codex, Cursor. Prompts get long, and typing all of that out every single time is genuinely tiring.
Just say what you are thinking. The text goes straight into your terminal too, so you can fire off the next instruction without breaking stride.
XAIV detects terminals and uses the paste method that actually works in them. Windows Terminal, WSL and Git Bash are supported.
Text about one second after you stop talking
Measured on roughly six seconds of speech (NVIDIA RTX 3060, small model). Your machine will differ.
Research says speaking is faster
A team from Stanford University and Baidu compared text entry on smartphones. Speech input came out 3.0× faster than the keyboard in English and 2.8× faster in Mandarin, with 20.4% and 63.4% fewer errors respectively.
Ruan, Wobbrock, Liou, Ng, Landay, “Speech Is 3x Faster than Typing for English and Mandarin Text Entry on Mobile Devices” (2016)
Note: that study compared smartphone keyboards, not desktop keyboards. Still, the longer the text you have to type, the more often speaking wins.
Speak the way you normally talk
Real speech is full of "um" and "uh", and people restart sentences. AI cleanup tidies that up, so you don't have to rewrite it yourself. Say what's in your head and get something readable.
This is off by default. Transcription works fine without it. Even when it is on, only text is sent to the AI — never audio.
⚠ Turning this on sends your transcribed text over the internet to an AI provider (the one you chose — it never passes through our servers). While it is off, transcription happens entirely inside your PC.
It learns your words as you use it
The proper nouns it always gets wrong, and the filler words you can't help saying — XAIV remembers them and fixes them automatically next time. Hold a dedicated key and just say "noshon to Notion" to register a correction, and it can strip fillers like "um" and "uh" on its own.
What it learns (correction rules, dictionary, fillers) is stored on your PC. Only text is used for learning — audio is never stored.
Personal learning is off by default. You can turn it on or off anytime in settings.
AI cleanup uses your own API key
AI cleanup is optional. To use it, enter your own API key. Supported: DeepSeek, Claude (Anthropic), ChatGPT (OpenAI), Gemini and KIE.ai. You pay the AI provider directly, and we add nothing on top.
Your API key is stored only on your PC and is sent only to the AI provider you chose. No usage data is sent to us. If you do not use AI, you need no key and pay nothing.
Pricing
- Unlimited transcription
- No sign-in, no account
- AI cleanup is optional (your own API key)
Only the AI provider's usage fee for AI cleanup is paid by you, directly. If you don't use it, it is zero.
System requirements
| OS | Windows 10 (64-bit) / Windows 11, macOS 10.15 or later (Apple Silicon and Intel), Linux (X11 session) |
|---|---|
| Memory | 8 GB or more (16 GB is comfortable) |
| Free space | 2 GB or more |
| Microphone | Built-in or external, either works |
| Graphics card | Not required. NVIDIA GPUs (Windows) and Apple Silicon Metal (Mac) are supported and make it several times faster |
| Internet | Not needed for transcription (your voice is never sent). Used for the first model download and update checks. If you turn on AI cleanup, it is also used to send the transcribed text. |
On Linux, use an X11 session (the hotkey does not work under Wayland).
Frequently asked questions
Do I need a microphone?
Yes. Most laptops have one built in, so you can start right away. On a desktop, use an external microphone or earphones with a mic.
Do I need an account with an AI service?
No. Transcription runs entirely on your PC, so it works without any AI account. Only if you want the text cleaned up (AI cleanup) do you enter an API key for DeepSeek, Claude, ChatGPT, Gemini or KIE.ai.
Could what I say leak out?
No. Your recording is turned into text on your PC and discarded as soon as that finishes. It is never sent to our servers.
Will it run on an older PC?
Yes. A graphics card is not required. If you have one, transcription is several times faster.
Which apps does it work with?
The text goes wherever your cursor is, so it works in almost anything — Word, Chrome, Slack, Notepad. Terminals and WSL are supported too.
Can I speak English or Chinese?
Yes. Sixteen languages are supported, including English, Chinese and Korean. The app interface is available in Japanese, English and Chinese.
Is there a Mac or Linux version?
Windows, macOS and Linux are supported. macOS 10.15 or later is supported (Apple Silicon or Intel). On Linux, use an X11 session — Wayland by design cannot detect the key press.
Does it cost anything?
No. Transcription is completely free, with no character limit and no account. Only if you use AI cleanup does the AI provider bill you for your own API key usage.
A world without typing
Get it freeCompletely free — no account needed