Openstyle 1.0.1

Openstyle is a local-first voice dictation app for macOS. Hold a hotkey, speak, release, and the text appears at your cursor.

Transcription runs on your machine. There is no account, no cloud service, and no telemetry.

Install

Download Openstyle-1.0.1.dmg (Apple Silicon) and drag it to Applications.

The build is ad-hoc signed, not notarized, so macOS will warn on first launch. Right-click the app and choose Open, or clear the quarantine flag:

xattr -cr /Applications/Openstyle.app

Openstyle needs Accessibility and Input Monitoring permission to read the global hotkey and paste at your cursor, and Microphone permission to hear you. macOS will prompt on first use.

What's changed

• (models) Connect to a self-hosted oMLX server for transcription.
• (cleanup) Make the cleanup model's sampling parameters editable.
• (logging) Write every model request and response to a local openstyle-trace.log for troubleshooting — always on, no setting to disable it, and it holds full transcripts and model output (only Authorization/api_key values are redacted). Never leaves the machine.
• (create-plugin) Rename the scaffolding CLI from create-freestyle-plugin to create-openstyle-plugin — the old name is published on npm at 0.0.1/0.0.2, switch to npx create-openstyle-plugin@latest. Starter templates now depend on the published @openstyle/sdk and the window.openstyle bridge instead of a stale freestyle reference.
• (server) Require auth and an allow-listed origin on the local API; redact credential fields from the bulk settings response.
• (server) Verify the MLX transcription-worker archive's checksum before extracting and running it.
• (sdk) Reject non-local import specifiers in the plugin loader, closing a remote-code-execution path. Plugin UI pages now get window.openstyle (was window.freestyle) and the manifest key is openstyle (was freestyle, still read as a fallback so installed plugins keep working).
• (db) Purge orphaned PostHog/cloud-sync rows left over from old local databases on migration to schema v28.
• (electron) Exclude workspace src, tsconfigs, and .turbo caches from the packaged app.asar; clean dist before pkgroll builds so stale output can't be packaged.
• Remove the mobile app from the repo; macOS desktop only going forward (never had a public release — no Android/iOS build ships from .craft.yml or the 1.0.0 release assets).

Upgrading from Freestyle

Openstyle uses its own application-support directory. On first launch it copies an existing Freestyle profile across, including transcription history, vocabulary and installed plugins. The original is left untouched, so nothing is lost either way.

Because it copies rather than moves, anything dictated in Freestyle after your first Openstyle launch stays in Freestyle.

Assets

• Openstyle-1.0.1.dmg — the app, Apple Silicon
• Openstyle-1.0.1-arm64.zip — same build, for the updater
• mlx_asr_worker-darwin-arm64.tar.gz — the on-device MLX speech runtime, fetched on first use of a local MLX model
• latest-mac.yml — update manifest

Known issues

• Auto-updating from 1.0.0 does not work: the app finds and downloads this update, but macOS rejects the ad-hoc code signature and the install fails. Download the .dmg from this page and install manually instead. A future version will skip the broken download and link here directly.

Openstyle 1.0.0

Openstyle is a local-first voice dictation app for macOS. Hold a hotkey, speak, release, and the text appears at your cursor.

Transcription runs on your machine. There is no account, no cloud service, and no telemetry.

Install

Download Openstyle-1.0.0.dmg (Apple Silicon) and drag it to Applications.

The build is ad-hoc signed, not notarized, so macOS will warn on first launch. Right-click the app and choose Open, or clear the quarantine flag:

xattr -cr /Applications/Openstyle.app

Openstyle needs Accessibility and Input Monitoring permission to read the global hotkey and paste at your cursor, and Microphone permission to hear you. macOS will prompt on first use.

About this fork

Openstyle is a fork of freestyle-voice/freestyle at tag 0.7.1, the last release before upstream removed on-device transcription and moved to a hosted, account-based product.

Removed: Freestyle Cloud, accounts and the sign-in gate, billing, PostHog telemetry.

Kept: on-device transcription with local MLX and whisper.cpp models, bring-your-own-key access to third-party providers, Remix, vocabulary, shortcuts, per-app tone, and the plugin system.

Thanks to the Freestyle authors and contributors for the original work. MIT licensed, as upstream is.

Upgrading from Freestyle

Openstyle uses its own application-support directory. On first launch it copies an existing Freestyle profile across, including transcription history, vocabulary and installed plugins. The original is left untouched, so nothing is lost either way.

Because it copies rather than moves, anything dictated in Freestyle after your first Openstyle launch stays in Freestyle.

Assets

• Openstyle-1.0.0.dmg — the app, Apple Silicon
• Openstyle-1.0.0-arm64.zip — same build, for the updater
• mlx_asr_worker-darwin-arm64.tar.gz — the on-device MLX speech runtime, fetched on first use of a local MLX model
• latest-mac.yml — update manifest