# Aqua Voice Overview

Aqua Voice is *AI-native* dictation for **Mac, Windows, and iOS**. Hold one key, talk into *any* text field — Cursor, Claude Code, Gmail, Slack, Notion, even your terminal — and your words are inserted, cleaned up and formatted for wherever you're typing. No per-app setup, no browser extension.

Aqua is powered by **Avalon**, our own speech-recognition model, paired with **Deep Context**, a client-side engine that reads what's on screen to nail names, code, and technical terms. It starts in under **50ms** and inserts text in about a second (as fast as ~450ms). At launch in October 2025, Avalon debuted at **#6 overall and #1 among proprietary models** on the independent [Open ASR Leaderboard](https://huggingface.co/spaces/hf-audio/open_asr_leaderboard) (6.24% WER average). In a [9to5Mac](https://9to5mac.com/2025/08/15/aqua-voice-shows-just-how-good-mac-dictation-could-be-if-apple-just-tried/) side-by-side against built-in macOS Dictation on the same passage, Apple made 17 errors; Aqua made 1.

## What is AI-native dictation?
It means two things:

1. **The AI that powers it** — Avalon (speech recognition) fused with language-model post-processing for accurate, contextual, well-formatted output, not just a raw transcript.
2. **The AI you talk to with it** — voice is the fastest way to prompt the tools you already use: Claude, ChatGPT, Cursor, Gemini, and any other AI app.

## What's New (as of August 2026)
- **Send It** (desktop v0.18.0, July 26, 2026): finish a hands-free dictation with "send it" and Aqua submits the message for you. Max plan, macOS and Windows, works in 6 languages.
- **Max plan** (July 2026): a new tier above Pro at $24/mo billed annually, covering live words with Realtime Mode, the Send It voice command, and early access to new features.
- **Edit Mode** (desktop v0.17.0, July 20, 2026): select text anywhere, hold your Aqua key, say the change, and Aqua rewrites the selection in place. Available to every Aqua desktop user, not a separate tier.
- **ISO/IEC 27001:2022 certification** (July 14, 2026), alongside the existing SOC 2 Type II.
- **Realtime mode**: the mode formerly called Streaming Mode was renamed in July 2026 and now sits on the Max plan.

## Platforms
- **macOS** — Apple Silicon and Intel (primary platform)
- **Windows** — 10 and 11
- **iOS 17+** — AI voice keyboard, launched April 2026; includes Voice Edit Mode (revise your last transcript by voice) and cross-device sync ([App Store](https://apps.apple.com/us/app/aqua-voice-ai-voice-keyboard/id6759074969))
- **Avalon API** — OpenAI SDK-compatible batch speech-to-text endpoint for developers

## Quick Start (3 min)
1. Download Aqua (links below) or get it on the [App Store](https://apps.apple.com/us/app/aqua-voice-ai-voice-keyboard/id6759074969).
2. Install and sign in.
3. Start dictation: hold **Fn** (Mac) or **Alt** (Windows); on iOS, switch to the Aqua keyboard and tap the mic.
4. Speak naturally — Aqua formats the text to match the app you're in.
5. Release the key (or tap again) to commit.

## Avalon (our model)
Avalon is Aqua's proprietary speech-recognition model, launched **August 2025** to replace our earlier Whisper-based pipeline. Unlike conventional ASR trained on audiobooks and news, Avalon is trained on **human-computer interaction** speech — prompts, code, and email — the way people actually talk to their computers.

- **Open ASR Leaderboard:** Avalon v1 debuted at **#6 overall, #1 proprietary** at 6.24% average WER on this third-party benchmark (October 2025); the live leaderboard moves as new models are added.
- **Avalon 1.5** (released April 2026 alongside iOS): **5.55% WER**, more than **2× faster** than Avalon 1, and wins **76%** of blind-test head-to-heads against ElevenLabs Scribe v2. Adds improved multilingual support.
- **AISpeak-10** (Aqua's own hard subset for AI/coding terminology, higher is better): Avalon **97.4%** vs ElevenLabs Scribe 78.8%, Whisper Large v3 65.1%, Voxtral Mini 59.5%, NVIDIA Canary 51.5%. (Avalon scores 97.3% on the full AISpeak set.)

Read more: [Introducing Avalon](https://aquavoice.com/blog/introducing-avalon) · [Avalon on the Open ASR Leaderboard](https://aquavoice.com/blog/avalon-openasr-leaderboard)

## Pricing
| Tier | Price | Highlights |
|---|---|---|
| **Starter** | Free | 1,000 free words; Avalon transcription model |
| **Pro** | **$8/mo** billed annually ($96/yr), or $10/mo month-to-month | Unlimited words; Custom Instructions; Expanded Custom Dictionary |
| **Max** | **$24/mo** billed annually, or $30/mo month-to-month | Everything in Pro; see words live with Realtime Mode; “Send it” voice command; early access to new features |
| **Team** | $12/user/mo billed annually, or $15/user/mo month-to-month | Everything in Pro; centralized billing; team-wide settings; enforce Privacy Mode |
| **Business** | Custom | SSO/SAML + SCIM; advanced reporting; Zero Data Retention; team-wide dictionaries; volume discounts |

- **Students:** 70% off Pro and Max with a .edu email, monthly or annual. Pro is $3/mo, or $2.40/mo billed annually; Max is $9/mo, or $7.20/mo billed annually.
- **iOS standalone:** $119/yr via the App Store, but a cross-platform Pro subscription already covers iOS.
- **Referrals:** share your in-app referral link and the person you invite gets a free month of Aqua; you get a $10 credit when they sign up.
- **What Max adds:** See words live with Realtime Mode, use the Send It voice command, and get early access to new features. Max includes everything in Pro.

## Developer API
The **Avalon API** supports a subset of OpenAI's batch transcription contract, so the official OpenAI Python and JavaScript SDKs work for supported requests by changing the base URL and model.

- Base URL: `https://api.aquavoice.com/v1`
- Transcription endpoint: `POST /v1/audio/transcriptions`
- Model: `avalon-v1.5` (multilingual)
- Pricing: **$0.39 per hour** of audio, billed per second with a 10-second minimum
- Docs: [aquavoice.com/avalon-api/docs](https://aquavoice.com/avalon-api/docs)

## Links
- [Try in Browser](https://aquavoice.com/sandbox)
- [Download](https://aquavoice.com/download)
- [Pricing](https://aquavoice.com/pricing)
- [Setup Guide](https://aquavoice.com/guide)
- [Blog](https://aquavoice.com/blog)
- [Changelog](https://aquavoice.com/changelog)
- [Aqua for Mac](https://aquavoice.com/mac)
- [Aqua for Windows](https://aquavoice.com/windows)
- [Realtime Mode](https://aquavoice.com/realtime)
- [Edit Mode](https://aquavoice.com/edit-mode)
- [Use Cases](https://aquavoice.com/use-cases)
- [FAQ](https://aquavoice.com/info/faq)
- [Privacy Policy](https://aquavoice.com/info/privacy)
- [System Status](https://status.aquavoice.com/)

## External Links
- [X / Twitter](https://x.com/aquavoice)
- [Aqua Voice on Y Combinator (W24)](https://www.ycombinator.com/companies/aqua-voice)
- [Hacker News — Launch (2024)](https://news.ycombinator.com/item?id=39828686)
- [Hacker News — Aqua Voice 2 (2025)](https://news.ycombinator.com/item?id=43634005)
- [Product Hunt](https://www.producthunt.com/products/aqua)

## Press & Independent Validation
- [9to5Mac — Aqua vs. Mac Dictation](https://9to5mac.com/2025/08/15/aqua-voice-shows-just-how-good-mac-dictation-could-be-if-apple-just-tried/) (17 Apple errors vs. 1 for Aqua)
- [9to5Mac — Aqua Voice on iPhone](https://9to5mac.com/2026/04/17/aqua-voice-the-best-dictation-app-ive-ever-used-is-now-available-on-iphone/)
- [Open ASR Leaderboard](https://huggingface.co/spaces/hf-audio/open_asr_leaderboard) — third-party WER ranking
- Product Hunt: 5.0/5; "Readability Award for AI Dictation Apps, Winter 2025"

## Tech Specs
### Downloads
Always-current installers live at **[aquavoice.com/download](https://aquavoice.com/download)** (auto-updating):
- macOS — Apple Silicon and Intel
- Windows 10/11
- iOS — [App Store](https://apps.apple.com/us/app/aqua-voice-ai-voice-keyboard/id6759074969)

### Dictation Languages (49)
Arabic, Belarusian, Bengali, Bulgarian, Cantonese, Catalan, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Irish, Italian, Japanese, Korean, Latvian, Lithuanian, Malay, Maltese, Mandarin, Marathi, Mongolian, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Slovenian, Spanish, Swahili, Swedish, Tamil, Thai, Turkish, Ukrainian, Urdu, Vietnamese, Welsh. Set the language picker to **Auto** to detect automatically.

### Latency
- Startup: under **50ms**
- Instant Mode: text appears ~**450ms** after you stop speaking
- Realtime mode: words appear in real time as you talk
- Avalon 1.5 is more than **2× faster** than Avalon 1

## Security & Compliance
Aqua Voice is **SOC 2 Type II certified** (completed **March 25, 2026**), demonstrating our commitment to the highest standards of security, availability, and confidentiality. The program was completed with [Vanta](https://www.vanta.com/); [Advantage Partners](https://advantage-partners.com/) was our auditor and [Workstreet](https://www.workstreet.com/) our advisor.

SOC 2 Type II means our security controls were independently audited over an extended observation period — not assessed at a single point in time — so teams and enterprises handling sensitive data can use Aqua with confidence. The report is available in our [Trust Center](https://app.vanta.com/withaqua.com/trust/6f0ehpcbakfh2e73cnoo3r). **Zero Data Retention** is available on Business.

Aqua Voice is also certified against **ISO/IEC 27001:2022** (announced **July 14, 2026**), the international standard for information security management. Certification means an accredited third party examined our Information Security Management System and confirmed the controls are designed and operating effectively. It is the framework security teams outside the United States most often ask for, which makes procurement and onboarding easier for global companies. Same auditor and partners as SOC 2: [Advantage Partners](https://advantage-partners.com/) auditing, [Vanta](https://www.vanta.com/) for evidence automation, [Workstreet](https://www.workstreet.com/) advising. The certificate is downloadable from the Trust Center.

Read more: [Aqua Voice is ISO 27001 certified](https://aquavoice.com/blog/iso-27001) · [SOC 2 Type II](https://aquavoice.com/blog/soc2)

### Privacy Mode and what happens to your data
Audio is processed in the cloud and is never sold, and it is never shared for advertising. With **Privacy Mode off** (the default), transcripts may be stored on our servers to improve the product. **Turn Privacy Mode on and they are not stored**; session metadata such as timestamps, device type, and performance metrics may still be collected. **Team** plans can enforce Privacy Mode across an organization, and **Business** adds Zero Data Retention. **Deep Context** (screen reading) is a separate setting, off until you enable it, and its data is never stored. **Instant Mode** dictation runs on Aqua's own Avalon models; **Realtime mode** streams audio to a third-party realtime speech provider. The current subprocessor list is in the [Trust Center](https://app.vanta.com/withaqua.com/trust/6f0ehpcbakfh2e73cnoo3r).

For security inquiries, contact **support@aquavoice.com**.

## Features
### Modes
Aqua has two operation modes:

- **Instant Mode** — Press key → talk → release → see text. Startup <200ms, results ~450ms. The default mode, available on every plan.
- **Realtime mode** — Press key → talk → see text in real time → release. Continuous output, best with Deep Context. Included on **Max**; it is what Send It and the other voice commands are built on. (Formerly called Streaming Mode; renamed in July 2026.)

Different apps call for different precision. You wouldn't sweat a Cursor prompt the way you would an important email. Instant Mode has the lowest latency and is great for short clips or "chaining" Aqua many times in a row; Realtime mode gives maximum contextual understanding and precise control over text and formatting. There's no right or wrong way — it's like texting style, some people send many short messages, others one long one.

### Deep Context
Deep Context uses what's on your screen to improve accuracy — especially for coding, messaging, and document editing. It's optional and off by default; data is processed securely and never stored.

Deep Context enables syntax-aware transcripts. For example:

```
You say:               Can you modify the canonical title on the context response model to be either an object or a string?
Aqua w/ Deep Context:  Can you modify the `canonical_title` on the `ContextResponse` model to be either an object or a string?
```

This works in any app but shines when prompting LLMs to generate code in Cursor, Windsurf, VS Code, Zed, JetBrains IDEs, and similar.

### Custom Instructions
Tune your output with natural-language rules, similar to ChatGPT custom instructions. For example:

```txt
In iMessage, Slack, & WhatsApp, use all lowercase (gen-z style) except for proper nouns and the pronoun "I".

# PUNCTUATION & SYMBOLS
- When I say "checkbox", insert a Markdown checkbox: - [ ]

# DICTIONARY & MACROS
- My Company: {{My Company Name}} (often misheard as {{common mispronunciation}})
- Our Product: {{Product Name}}
```

See our Discord for recommended formats and starter packs.

### Dictionary
'Impossible' words are no problem. Add names, technical terms (YC W24, factorio, package.json), or custom phrases (Jeffersonian Computing, OPEN-RISOP) and Aqua recognizes them in speech while preserving casing. It's context-aware too: enter "factorio" and Aqua still capitalizes it correctly in "Factorio is a great game."

### Edit Mode
Select any text, in any app, hold your Aqua key, and say the change. Aqua rewrites the selection in place. No mouse, no retyping, no second hotkey to learn. Hold the key with nothing selected and you get normal dictation, exactly as before; hold it with a selection and Aqua switches into Edit Mode automatically.

Live now for every Aqua desktop user on **macOS and Windows** (shipped in v0.17.0), not a separate purchase or tier. Works in native text fields, browsers, Google Docs, chat and email clients, code editors, note apps, and rich-text editors. Wherever you can select text, you can talk to it. No command words needed: say the corrected version of what you selected and Aqua treats it as the replacement. Edits stack and undo ("undo that," "go back one step," "go back to the original").

iOS has its own **Voice Edit Mode** for revising your last transcript by voice (see iOS platform notes above); the desktop version is selection-based and works across any app you use.

Read more: [Introducing Edit Mode](https://aquavoice.com/blog/introducing-edit-mode)

### Send It
End a hands-free dictation with "send it" and Aqua strips the command out of your text, inserts the message, and presses Return for you. No reaching back for the keyboard to send.

Shipped in desktop **v0.18.0** (July 26, 2026) on **macOS and Windows**, included on **Max**. There is no Send It setting and no second hotkey: it runs on Realtime dictation, using the same key you already hold for Edit Mode. Start a hands-free session by double-pressing your Aqua key (Fn on Mac, right Alt on Windows) or by holding it and tapping Space, then say your message ending in "send it." The command itself ends the session.

It is not an integration with any one app. Aqua inserts at the cursor system-wide, so Send It fires anywhere Return would submit: chat apps, message threads, browser chat boxes, and the prompt boxes of AI agents.

Aqua reads the whole utterance rather than matching a keyword, so the command has to be the last thing you say. "Running late, be there soon, send it" sends; "Tell them I'll send it tomorrow" stays as text. You do not need an exact phrase: "send," "send message," and polite forms like "please send this message" all work.

Send It is not English-only. Natural commands work in Japanese (送って, 送信してください), Spanish (envíalo, envía el mensaje), French (envoie-le, envoie le message), German (schick es ab, abschicken), and Russian (отправь, отправь сообщение). English "send it" is honored in every language, so it keeps working if you switch languages during the day.

Read more: [Introducing Send It](https://aquavoice.com/blog/introducing-send-it)

### Computer Control (beta, macOS)
An AI agent that operates your cursor and keyboard by voice. Triple-tap to start, say what you want in plain language, and it takes the actions; you can steer it mid-task, set guardrails (asking before it sends a message or deletes anything), or stop it at any point. Unlike classic voice-control tools there is no fixed command grammar to memorize.

Beta, macOS only, included on **Max** and **Business**.

### File Tagging
Say a filename while dictating in **Cursor or Windsurf** and Aqua converts it into a real editor tag that triggers autocomplete. "main ts" becomes `@main.ts`, "plates events ts" becomes `@plates.events.ts`, "main dot pi" becomes `@main.py`. You can signal a filename explicitly with a trigger word ("at main", "tag user profile"). When a project has many files with the same name (page.tsx, index.ts, Dockerfile), the tag hands off to the editor's own autocomplete so you pick the right one. Supported in Aqua v0.10 and up. See [aquavoice.com/guide/file-tagging](https://aquavoice.com/guide/file-tagging).

### Replacements
Short spoken phrases that expand into longer text: say "my calendar link" and Aqua inserts the URL. Useful for LLM prompts you repeat, email addresses, and boilerplate. Add them in the Replacements tab. Distinct from the Dictionary (which teaches Aqua to recognize hard words) and from Custom Instructions (which shape style). Supported in v0.10.8 and up. See [aquavoice.com/guide/replacements](https://aquavoice.com/guide/replacements).

### Additional Features
- **Language autodetect** — select "Auto" in the language picker.
- **Local History** — transcripts and audio are kept locally so you never lose text; see the "History" tab in settings.
- **Multiple hotkeys** — bind more than one key to activate Aqua (handy across keyboards or to avoid conflicts).

## Where Aqua Works
Aqua is system-wide, not a browser extension or a per-app plugin: it inserts at the cursor anywhere there is a text field. There are per-app pages at `aquavoice.com/use-cases/<name>` covering setup and behavior for:

- **AI tools:** antigravity, chatgpt, claude, codex, cursor, gemini, perplexity, vscode
- **Messaging:** discord, imessage, line, slack, telegram, whatsapp
- **Terminals:** ghostty, iterm, terminal, warp
- **Email:** applemail, gmail, outlook, superhuman
- **Docs and notes:** arc, googledocs, linear, excel, word, notion, obsidian

## How Aqua compares
Factual, sourced positioning — for full benchmarks see the Avalon links above.

- **vs. Wispr Flow** — Aqua is cheaper ($8 vs. $15/mo) and runs on its own proprietary model (Avalon) with published third-party benchmarks; Wispr offers more languages (100+) and Android.
- **vs. Willow Voice**: Aqua runs its own proprietary model (Avalon) with published third-party benchmarks, while Willow does not publicly disclose its model; Aqua is cheaper ($8 vs. $15/mo); Willow offers more languages (100+) and a HIPAA Ready posture. Full comparison: [aquavoice.com/vs/willow](https://aquavoice.com/vs/willow).
- **vs. Superwhisper** — Aqua is cloud-based with screen context for accuracy; Superwhisper is on-device by default (local models) and offers a lifetime plan (now on Mac, Windows, and iOS).
- **vs. Apple Dictation** — independent [9to5Mac](https://9to5mac.com/2025/08/15/aqua-voice-shows-just-how-good-mac-dictation-could-be-if-apple-just-tried/) test: 17 errors vs. Aqua's 1 on the same passage.
- **vs. Whisper / cloud ASR APIs** — Avalon debuted as the #1 proprietary model on the Open ASR Leaderboard (October 2025) and leads on AI/coding terminology (AISpeak).

Prices above compare Aqua **Pro** ($8/mo annual) against each competitor's individual plan. Full side-by-side pages live at `aquavoice.com/vs/<name>` for: betterdictation, dragon, handy, lemon, loqua, mac-dictation, monologue, openai-whisper, openwhispr, paraspeech, resonant, speechpulse, spokenly, superwhisper, typeless, voibe, voicedash, voiceink, voicy, willow, wispr-flow.

## Company
Aqua Voice is a San Francisco company founded in **2023**, part of **Y Combinator's Winter 2024 batch**. Founders: **Finn Brown** (CEO) and **Jack McIntire** (CTO). Tagline: *"We've typed for 150 years. It's time to speak."*

## Feedback
Have suggestions or want to share results? Email <support@aquavoice.com> or join our [Discord](https://discord.gg/aqua-voice).

*(Updated August 16, 2026)*