Handy v0.10.0 transcribes audio files and detects your language with every model

Handy now transcribes audio files as well as your voice. Click Transcribe File in History, or drag a file onto Handy's window, and the text lands in your history.

Slide text

01 · FILES

Handy v0.10.0 transcribes audio files and detects your language with every model

Source: cjpais/Handy · src/content/release-notes/0.10.0.md

Long audio is now cut into chunks by default, with voice detection finding the breaks. That speeds up long transcriptions and stops some models from crashing. You can switch it off under Advanced.

Slide text

01 · FILES

Long recordings are split into chunks before they’re transcribed

On by default. Switch it off in Advanced → Long Audio Chunking.

Source: cjpais/Handy · src/content/release-notes/0.10.0.md

Getting started, and picking a model.

First-time users get a new onboarding. It walks through permissions, picking a model and a first try at dictation, so a speech model on your own computer is easier to start with.

Slide text

02 · MODELS

New users get a rebuilt onboarding for local models

Welcome Permissions Model Try it

Source: cjpais/Handy · release-notes/0.10.0.md · src/components/onboarding

Tell Handy which languages you speak, under Preferred Languages. The rebuilt model picker grades every model in each of those languages, and marks one Recommended, one Faster and one More accurate.

Slide text

02 · MODELS

Every model gets a grade from A to F in each of your languages

Set them in General → Preferred Languages.

Source: cjpais/Handy · src/content/release-notes/0.10.0.md

Auto language detection now works with every model. Models that can't tell the language use a small detector that downloads when you pick Auto. An experimental setting limits Auto to your preferred languages.

Slide text

02 · MODELS

Auto language detection works with every model

Models that can’t tell the language use a small detector, downloaded when you pick Auto.

Source: cjpais/Handy · src/content/release-notes/0.10.0.md

You can switch the dictation language from the menu bar, without opening settings. Auto detect sits at the top, followed by up to five languages you used recently.

Slide text

02 · MODELS

Switch the dictation language from the menu bar

• Auto detect always first
• Recent your last five languages
• The label names the active language

Source: cjpais/Handy · GitHub release v0.10.0 · PR #1319

Custom words are the names and terms you add so Handy spells them right. They used to help only Whisper. Now Qwen3 ASR, Granite and other models get them directly.

Slide text

02 · MODELS

Custom words now reach Qwen3 ASR, Granite and more

Before this release, only Whisper models used them.

Source: cjpais/Handy · src/content/release-notes/0.10.0.md

When something goes wrong.

Transcription now runs in a separate process. If the model or your GPU crashes or hangs, Handy keeps running and recovers. When the GPU was the cause, it retries on the CPU.

Slide text

03 · STABILITY

Transcription runs in its own process, so a model crash leaves Handy running

If the GPU caused it, Handy retries on the CPU.

Source: cjpais/Handy · src/content/release-notes/0.10.0.md

Errors now appear in the recording overlay right away, and in a tray inside the app. A model that's still loading says so. And if post-processing comes back blank, Handy keeps your dictation.

Slide text

03 · STABILITY

Errors show up the moment they happen

• Errors in the overlay and in a tray in the app
• Model loading shown instead of a stuck “Transcribing…”
• Blank post-processing keeps your original dictation

Source: cjpais/Handy · GitHub release v0.10.0 · PRs #2372, #2227, #2290

On macOS, AirPods could give Handy a silent recording while your iPhone played audio through them. The experimental AirPods Mode fixes that, and the iPhone's playback stops when you start recording, as it does in Voice Memos.

Slide text

03 · STABILITY

AirPods Mode stops silent recordings while your iPhone plays audio

macOS · experimental · iPhone playback stops when you start recording

Source: cjpais/Handy · src/content/release-notes/0.10.0.md · PR #1991

For power users.

On macOS and Linux, Handy can pass every transcription through a script of your own before pasting. Whatever the script prints is what gets pasted. If it fails or runs past 30 seconds, the original text goes in.

Slide text

04 · POWER USERS

A transcription hook runs your own script before Handy pastes

macOS and Linux · the script’s output is what gets pasted

Source: cjpais/Handy · README.md, Transcription hook · PR #930

If a language model cleans up your dictation, the Custom provider now speaks Anthropic's Messages API, as well as OpenAI's format. So Claude works directly, even in Microsoft Foundry, without a proxy like LiteLLM.

Slide text

04 · POWER USERS

The Custom post-processing provider speaks Anthropic’s Messages API

Claude, including Claude in Microsoft Foundry, with no proxy in between.

Source: cjpais/Handy · GitHub release v0.10.0 · PR #2257

What's gone.

Handy no longer runs ONNX models. If you still use one, such as Parakeet V3, Moonshine or Canary, Handy offers its GGUF version and guides you through the switch. It should run as well on your machine, or better.

Slide text

05 · REMOVED

ONNX models are gone, replaced by GGUF versions

Parakeet V2 Parakeet V3 Moonshine SenseVoice GigaAM v3 Canary Cohere

Handy offers the GGUF replacement when it starts.

Source: cjpais/Handy · src/content/release-notes/0.10.0.md · src-tauri/src/managers/model.rs

This release makes Handy roughly feature complete, and the next ones fix issues and polish the app on the way to version one. For dictating on your own computer, Handy is what I'd use.

Slide text

Next for Handy: stability and polish on the way to V1

Full list: Handy’s release notes on GitHub, and What’s New in the app.

Source: github.com/cjpais/Handy/releases/tag/v0.10.0