Tildo — Voice to text on your Mac. Local. Open source.

01 · See it

Press fn, talk, release.

A pill appears while you speak. When you let go, the transcript lands in the app you were using — cleaned up if you have a tone enabled.

New Message · To: team@tildo.app

you said fn

hey team quick heads up the feature branch is finally green so we can probably merge tomorrow morning if nothing blows up overnight

Tildo wrote

Hi team,

Quick heads-up — the feature branch is finally green. Assuming nothing breaks overnight, we should be able to merge tomorrow morning.

— Rodrigo

what you said what Tildo wrote tone: Professional

02 · How it works

Four things, in order.

No magic. Just a pipeline that runs on your machine.

Hold fn

A small pill shows up in the corner. Doesn't steal focus, doesn't interrupt.

Speak

Audio captured locally. Whisper.cpp on your Mac — faster than real-time on Apple Silicon.

Release

Transcription finishes. If a tone is on, an LLM rewrites the raw text to match.

Paste

Result lands in whatever app you were using. Mail, Slack, Xcode, Notion, the terminal.

03 · Tones & LLM

Rewrite what you said, your way.

Four tones ship by default. Write your own. The LLM is optional — when off, Tildo just pastes the raw transcript.

so uh yeah i think we should probably push the launch to next week because the api thing isn't ready and i don't want to ship something broken

Professionalbuilt-in

I'd suggest pushing the launch to next week. The API isn't ready and I'd rather not ship something broken.

Claude

Casualbuilt-in

Think we should push launch to next week — API's not ready and I don't want to ship something broken.

Claude

Bullet notesbuilt-in

• Push launch to next week
• API isn't ready
• Avoid shipping broken

Claude

Rawno LLM

so uh yeah i think we should probably push the launch to next week because the api thing isn't ready and i don't want to ship something broken

—

Plug in your own key: Anthropic OpenAI Claude Code CLI keys stored in Keychain

04 · Whisper models

Pick a size. Download once.

All Whisper models run locally. Bigger is more accurate, smaller is faster.

Model

What it's good for

Size

Speed

Accuracy

Tiny

Quick notes, single-language, low-power Macs

39 MB

Base

Balanced default. Great on any modern Mac.

142 MB

Small

Solid accuracy for mixed-language dictation

465 MB

Medium

Long-form dictation, technical terms, multilingual

1.5 GB

Large-v3

Best accuracy. Heavy. For M2 Pro and up.

2.9 GB

05 · Privacy

Yours stays yours.

Tildo is built so the default case — you talking to your Mac — never touches a server.

No account, no sign-inInstall and use. Nothing to log into. No email, no password, no ID.

No analytics, no telemetryTildo doesn't phone home. No events, no crash pings, no usage counters. The code is on GitHub — check.

Your voice never leaves the MacWhisper runs locally. Audio is processed in memory and dropped. Nothing saved unless you ask.

LLM is opt-in and yoursIf you turn on a tone, Tildo uses the provider whose key you configured. Off by default. Your key, your account.

Open source, MITEvery line is auditable. Fork it. Package it. We don't know, because we can't.

Your voice captured locally · never saved

            
            Whisper.cpp
            transcription · on your Mac · no network

Pasted into your app Mail, Slack, Xcode, Notes — anywhere

internet · optional

LLM tone rewrite text only · your key · off by default

opt-in

Talk to your Mac.
It writes back.

Press fn, talk, release.

Four things, in order.

Hold fn

Speak

Release

Paste

Rewrite what you said, your way.

Pick a size. Download once.

Yours stays yours.

Less typing. More thinking.