Fast voice transcription for AI builders

Talk to Claude,
don't type.

Fast local voice transcription,
no subscription.

Doing uses an on-device AI model to transcribe your voice in real-time. Hold fn, talk, release — your words land wherever your cursor is. $49 once.

macOS 14+ · Apple Silicon & Intel · No account required

fnHold fn to record
YOLO Mode ↵

add a dark mode toggle to the settings page

2:23pm to Claude Code

the signup button isn't showing on mobile, can you fix it

2:21pm to Cursor

brainstorm with me some new ideas for how we can improve the speed and accuracy of transcription

2:18pm to Claude Code

Hold fn Talk. Done.

1

Hold your hotkey

Press and hold your hotkey (fn by default). Doing starts listening instantly. A small pip appears at your cursor so you know you're live.

2

Speak

Talk naturally. Say what you'd type. System volume lowers automatically so you stay focused. Doing transcribes in real-time.

3

Release

Let go of fn. Your words are pasted wherever your cursor is — Claude Code, ChatGPT, Cursor, anywhere.

Own your tools, stop renting.

Every other voice tool charges $8–15 a month, forever. That's another subscription you have to think about, another thing to cancel if you stop using it. Doing is $49 once. It's yours. No recurring charge. No account. No free tier with artificial limits. Just a tool you own.

Doing

$49 once

Year 2 cost: $0

Runs locally

No account needed

Yours forever

Other voice tools

$8–15/mo

$96–180 per year

Cloud-dependent

Account required

Cancel and it's gone

Download for Mac— $49

3 device activations · 14-day refund guarantee · Free updates

Transcription with zero wait.

Doing's default engine transcribes locally in real-time — no upload, no server, no spinner. By the time you release fn, your words are already there.

150xrealtime

60 seconds of audio transcribed in ~400 milliseconds. On-device.

Parakeetlocal
150x
Applelocal
25x
Geminicloud
23x
OpenAIcloud
22x
AssemblyAIcloud
20x

Measured using Doing's built-in benchmark tool. Cloud providers available via bring-your-own-key.

Your voice never leaves your Mac.

No audio uploaded. No server processes your speech. No cloud account stores your data. Everything happens on your machine.

This isn't a privacy policy. It's architecture.

Audio

Processed in real-time

Your voice is transcribed as you speak, then discarded. Never recorded. Never stored. Never uploaded.

Local AI Models

Runs on your hardware

Parakeet runs entirely on your Mac. Apple's foundation model too, when available. No internet. No API calls. Works on a plane.

Transcripts

Daily markdown files

Every transcription saved locally. Searchable. Obsidian-compatible. Yours to keep.

## 2:23pm
make the onboarding flow
three steps instead of five

Want cloud? Bring your own API key for OpenAI, Gemini, or AssemblyAI. You choose when. You pay only for what you use.

Works with every AI tool and app.

Doing works at the system level. Anywhere you can type, you can talk. Hold fn in your browser, editor, or terminal — Doing transcribes and pastes wherever your cursor is.

✳ Claude Code
myappgit:(main)claude

Claude Code v2.1.45

Opus 4.6 (1M context) · ~/doing/myapp

myapp on main | Opus 4.6 (1M context)
Claude CodeChatGPTClaude.aiCursorCodexWindsurfSlackNotionGmailAny text field

Always know you're live.

When you hold fn, a small pip appears at your cursor. It follows your mouse so you always know two things: that Doing is listening, and exactly where your words will land.

The waveform inside reacts to your voice in real-time. When you stop talking, it settles. When you release fn, it disappears. No chrome, no window, no distraction.

Try it — hold Shift and move your mouse.

YOLO Mode

Auto-submit your transcriptions. No reviewing, no pressing Enter.

Release fn and your words are pasted and sent. LLMs understand what you mean, imperfections and all — just say what you're thinking and let the AI figure it out.

YOLO Mode

Auto-press Return after pasting transcription

No slow AI middleman.

Other voice tools run your speech through an LLM to “clean it up” before giving it to you. That adds seconds of processing to every single prompt — and rewrites what you actually said.

Doing skips that step entirely. You're already sending your words to an AI. Why run them through another AI first?

Doing

speak → transcribe → paste

That's it.

Other voice tools

speak → transcribe →

LLM rewrite (1-3s) →

paste

Want AI cleanup for emails or Slack? Doing has Skills for that. But the default is your words, instantly, with nothing in between.

5 engines. 1 benchmark tool.

Choose your engine with data, not marketing.

Doing ships with Parakeet — an on-device model that transcribes 15 seconds of audio in 180 milliseconds. But we don't lock you in. Bring your own API key and benchmark every engine side-by-side on the same audio.

Local · free forever: Parakeet (v3), Apple Foundation

Cloud · bring your key: OpenAI Whisper, Gemini, AssemblyAI

Cloud engines use your own API key — you pay the provider directly, per use. No middleman markup.

Benchmark Results
15.0s audio5 providers
Parakeet (v3)localactivefastest
Time0.10s
Speed150x realtime
Costfree

Okay, we're gonna test the benchmarking tool. I'm just speaking out loud here while um to collect some words for the transcription process.

Download Doing to run your own benchmarks

Built by a builder

I built the tool I needed.

I'm Brian. I use Claude Code for hours every day — over a hundred voice prompts a day, every day. I was tired of typing all of them. And I didn't love that every voice tool on the market was sending my audio and transcripts to their servers, doing who knows what with my data.

So I built Doing. Now it's how I build everything — including Doing itself.

No VC funding. No growth targets. No pressure to become a platform. Just a tool that does one thing well.

Local. Fast. Yours.

Voice input for AI builders. No cloud. No subscription. $49 once.

Download for Macv0.1.0

macOS 14+ · Apple Silicon & Intel · No account required