Entune 0.3.0

Entune now looks and feels like a Mac app, builds your dictionary in the background while you keep dictating, and finds more of the words your speech model gets wrong.

uv tool install entune && entune

In Windows PowerShell: uv tool install entune; entune. To upgrade: uv tool upgrade entune && entune.

What's new

• At home on macOS. The window uses the colours macOS itself uses, in dark and light, with the system blue accent. Its side rails and toolbar show the Mac's translucent window material, as in Finder and System Settings. The recording pill and the first-launch introduction follow your theme.
• Dictation keeps working while the dictionary learns. Suggestions run in the background; only editing the dictionary waits until you review them.
• Better suggestions. The suggestion model now sees only the dictionary entries that occur in the transcripts it reads, in a compact form, so requests are less than half the size. Its main job is finding words the dictionary does not cover yet. You choose the suggestion model and its reasoning effort, from minimal to xhigh; high is the default.
• Reuse earlier transcripts. Each transcript is read once by suggestions you apply. To read them again, for example after clearing your dictionary or choosing a stronger model, turn on Reuse earlier transcripts and pick a span on the timeline. Help next to it shows what has been read so far.
• Faster dictionary generation from audio. Imported audio is transcribed four recordings at a time with a cloud speech model, and suggestions start as soon as enough text is ready. The setup estimates the time from what Entune has measured on your Mac.
• Clear recovery. A reply that runs too long, for example past the ChatGPT plan's limit of about 15 minutes, stops the run instead of retrying for up to 45 minutes. Finished parts are kept, and Continue picks up where it stopped, with a lower reasoning effort if you choose.
• Optional tracing. Save Langfuse keys under Settings › Integrations to see every request and reply of the suggestion model. Off unless you add keys; install with uv tool install --force "entune[tracing]".
• Drop audio to transcribe it. Drop audio files anywhere on the window, such as a recording another dictation app could not transcribe. Each one is transcribed with your default model and appears in History.
• Anonymous mode. The eye switch in the toolbar blurs your transcripts for screen recordings. The newest stays readable until you copy it.
• Get a key. Each cloud provider, and TypeSafe for Jev, links to its API-key page.
• Stop from the window. While a shortcut recording runs, the window's button reads Stop and ends it.

Fixes

• A microphone that hangs while closing no longer loses the dictation or freezes the shortcuts and Quit. The audio is kept and transcribed, and the next recording asks you to reopen Entune.
• A suggestion reply is no longer sent back because a heard form differed from its spelling only in capitals.
• Other websites can no longer show Entune's page in a frame or load its recordings.
• On Arch Linux, a missing system library is named with the pacman command that installs it.
• The Dictionary page has a visible Help after Add, and the help in the suggestions panel covers only what that panel does. The Dictionary filters and the Settings layout are clearer.

Tested by hand on macOS. Windows and Linux were covered by the automated tests, not by hand, for this release.

Entune 0.2.1

Entune now runs as its own app on Linux too, and the first run is smoother everywhere.

uv tool install entune && entune

In Windows PowerShell: uv tool install entune; entune. To upgrade: uv tool upgrade entune && entune.

What's new

• Linux desktop app. entune adds Entune to the applications menu and opens it as its own window, with a tray item, global shortcuts and paste into the app you are in, on X11 and Wayland. Get started shows the one-time keyboard setup command, and entune names any missing system library with the command that installs it.
• No browser. Entune is its own app on macOS, Windows and Linux; nothing opens a browser.
• A first-launch introduction. The first time, Entune opens through a short introduction with the four setup steps, then Get started.
• The pill tells you what happened. Level bars while you record; then "Pasted", or why the text was only copied. An error stays on the pill with the provider's message and Retry, which transcribes the same recording again. Results no longer raise system notifications.
• Your clipboard is left alone. A paste borrows the clipboard and puts back what was there (on Windows, copied text only).
• Model downloads can be cancelled.

Fixes

• macOS asks for Microphone, Accessibility, then Input Monitoring, so its one Quit & Reopen comes last.
• The first Fn hold after setup stops when released.
• Pasting no longer stops after a while because of a missed key release.
• Learning from audio skips a recording that will not transcribe (after one more try) instead of stopping, and Retry continues where it stopped.
• Silence: nothing heard at the start shows on the pill at once; a pause after speaking is tolerated for about 40 seconds.
• A Wispr Flow import that macOS refuses now says so instead of "no database found".
• The Models page opens on its first section.

Tested by hand on macOS 26 and Windows. Linux was tested on an Ubuntu 24.04 desktop image (install, window, keyboard setup, introduction) but not yet on real hardware with Wayland.

Entune 0.1.0

Entune 0.1.0 is a local dictation app with a choice of speech providers and a personal dictionary that selects corrections in context.

• Record, replay, transcribe, retry and copy in the browser; keep recordings and history locally.
• Use cloud speech providers or separately installed local engines. macOS also supports global dictation shortcuts and paste.
• Build a dictionary from history or imported audio, review its suggestions, and choose Jev or local Laya for contextual replacements.
• Dictionary generation can use a provider API key or the experimental ChatGPT subscription sign-in option. Generation takes minutes to hours for large histories; audio selection shows a rough estimate.

Install with uv tool install entune, or python -m pip install entune in a Python 3.12+ environment. Then run entune and select a speech model. See the README and model guide for provider keys, engine setup and macOS permissions.

Windows and Linux browser workflows have not been tested end to end; native shortcuts and paste require macOS. Entune assumes a trusted single-user machine: its loopback API has no authentication and must not be exposed to a network or tunnel. Cloud speech and decision services receive their configured inputs; local history is not hosted by Entune.