Tildo

Speak. It types. A macOS menu bar app that turns your voice into text — right where your cursor is. No cloud, no subscriptions, no data leaving your Mac.

Built on OpenAI's Whisper running locally via whisper.cpp.

---

Why Tildo?

• 100% private — Everything runs on your Mac. Your audio never leaves the device.
• Works everywhere — Text appears wherever your cursor is: emails, code editors, Slack, browsers, terminal. Any app.
• One hotkey — Press your shortcut, talk, done. No window switching, no copy-pasting.
• Actually fast — Whisper runs natively on Apple Silicon. A 30-second recording transcribes in under 2 seconds with the right model.
• Free and open source — No subscriptions, no monthly fees. LLM post-processing is optional and uses your own API keys.

Features

Two ways to dictate

• Batch — Record everything, then transcribe at once. Great for long thoughts.
• Live — See your words appear in real-time as you speak.

16 languages Auto-detects or you choose: English, Spanish, French, German, Italian, Portuguese, Chinese, Japanese, Korean, Russian, Arabic, Hindi, Dutch, Polish, Turkish.

Pick your model 21 Whisper models from Tiny (32 MB) to Large v3 Turbo (1.6 GB). Download and switch from within the app. Quantized versions (Q5, Q8) give you a good balance of speed and accuracy.

LLM post-processing Optionally pass transcribed text through an AI model to correct, reformat, or translate. Works with Ollama (fully local, no API key), Claude Code, OpenAI, Anthropic, or Groq.

Per-app tones Define custom AI instructions per application — formal in email, casual in Slack, technical in your IDE. URL-pattern matching for browsers.

Text replacements Substitutions applied after transcription: "arroba" → "@", "hashtag" → "#", or anything you define.

Keyboard shortcuts Assign any key combination for record and cancel-recording from Settings → Atajos. Live key-cap feedback while pressing, 3-second hold-to-confirm.

And more

• Auto-stops when you go silent (configurable timeout)
• Sound feedback on start and stop
• Searchable transcription history

Installation

1. Download the latest Tildo-x.x.x-arm64.zip from the Releases page
2. Unzip and drag Tildo.app to your /Applications folder

Step 1 — Allow the app to run (required, one time only)

Tildo isn't notarized, so macOS will block it on first launch. Pick the method for your macOS version:

macOS 15 Sequoia

1. Double-click Tildo.app — macOS shows a "cannot be opened" dialog. Click OK.
2. Open System Settings → Privacy & Security, scroll down.
3. You'll see "Tildo was blocked" — click Open Anyway, then Open to confirm.

macOS 14 Sonoma or earlier

Right-click (or Control-click) Tildo.app → Open → click Open in the dialog. Done.

Any version — Terminal (fastest)

xattr -cr /Applications/Tildo.app

Then double-click the app normally. You only need to do this once.

Step 2 — Grant permissions (required, one time only)

Tildo needs two permissions to work:

Microphone — macOS will ask automatically on first launch. Click Allow.

Accessibility — needed to type text into other apps.

1. Open System Settings → Privacy & Security → Accessibility
2. Toggle Tildo on

Without Accessibility, Tildo still transcribes — the text just won't be typed automatically. You can paste it manually with ⌘V (set Output to Clipboard in Settings).

Local LLM with Ollama

Tildo can use any model running locally via Ollama — no API key, no internet, fully private.

Setup

1. Install Ollama from ollama.com

2. Pull a model. Recommended for Apple Silicon:

   ollama pull gemma4:e4b   # ~3 GB, best quality/size balance
   ollama pull gemma4:e2b   # ~1.5 GB, lighter option
   ollama pull qwen3:4b     # ~2.4 GB, strong multilingual

3. In Tildo, go to Settings → LLM, and click Detect on the Ollama card.

Ollama runs on localhost:11434. Tildo connects automatically — as long as Ollama is running in the background, everything works offline.

Tip: run ollama list to see which models you have installed, then type the model name directly in the model field.

Build from source

1. Build the whisper.cpp framework

git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp
cmake -B build -DBUILD_SHARED_LIBS=ON -DCMAKE_OSX_ARCHITECTURES="arm64"
cmake --build build --config Release

Package it and place it in Frameworks/:

xcodebuild -create-xcframework \
  -library build/src/libwhisper.dylib \
  -headers include/ \
  -output Frameworks/whisper.xcframework

2. Build

bash build.sh
open Tildo.app

Or open Package.swift in Xcode.

3. Download a model

On first launch, go to Settings → Modelos:

Goal	Model	Size
Fastest	Base Q5	60 MB
Balanced	Small Q5	190 MB
Best accuracy	Large v3 Turbo Q5	574 MB

Built with

• whisper.cpp — C/C++ port of OpenAI Whisper
• SwiftUI — Native macOS UI
• AVFoundation — Audio capture
• CoreGraphics — Keystroke simulation

License

MIT