Skip to content

Fixer 0.2.0 — Gemini Dictation

Choose a tag to compare

View all tags
@iliawerner iliawerner released this 20 Aug 05:35
· 12 commits to main since this release
v0.2.0
6241a35
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Fixer 0.2.0 is the second public beta. The headline feature is
Gemini-powered Dictation, accompanied by a complete redesign of the app as a
compact native macOS utility.

What's new

Gemini Dictation

  • A permanent, protected Dictation Action records speech and inserts the
    transcript at the original cursor.
  • Choose Press again to start and stop with two Shortcut presses, or
    Hold for hold-to-talk.
  • Language recognition is automatic and asks Gemini to preserve mixed-language
    speech without translating it.
  • Add {voice} to any ordinary Action to combine spoken instructions with
    optional {text}, then process the result with that Action's selected model.
  • If the original app, field, selection, or cursor changes while Fixer is
    working, the result is copied to the clipboard instead of being pasted into
    the wrong place.

New native design

  • A smaller, focused Actions workspace with one-column settings and a compact
    sidebar.
  • A warm paper and signal-yellow visual system with responsive hover, press, and
    focus states.
  • A redesigned passive status card for listening, processing, success, and
    errors that never takes focus from the active app.
  • A layered floating splash card with pointer depth, highlights, and Reduce
    Motion support.
  • Clearer provider Setup and Starter Library flows.

Privacy and reliability

  • The Gemini API key remains in macOS Keychain.
  • Microphone permission is requested only when a voice Shortcut is first used.
  • Voice recordings are held only in memory, limited to five minutes, and sent to
    Google Gemini gemini-3.7-flash for transcription.
  • Fixer does not save recordings or transcript history.
  • Escape cancels while audio is still local. Once Transcribing begins, the
    request may already have been uploaded.
  • Clipboard restoration, Shortcut coordination, Keychain error reporting, and
    stale provider-response handling have been strengthened.

Updating from 0.1.0

Quit the old Fixer build, replace it in Applications, and launch 0.2.0.

Existing Actions, Shortcuts, and the Gemini key are preserved. Fixer adds the
new Dictation Action at the top of the list; assign it a Shortcut to use voice.

macOS may ask for Accessibility again after an ad-hoc signed build is replaced.
Microphone access is requested separately only when Dictation or a {voice}
Action is first invoked.

Requirements

  • macOS 13 Ventura or later
  • Apple Silicon or Intel Mac
  • A personal Gemini API key and internet connection
  • Accessibility permission
  • Microphone permission for voice features only

Installation note

This build is ad-hoc signed and not notarized. First Control-click Fixer.app
and choose Open. If macOS still blocks it, use System Settings → Privacy &
Security → Open Anyway. Do not disable Gatekeeper globally.

If Finder specifically reports that the app is damaged, verify
SHA256SUMS.txt from this release and only then remove quarantine from this app:

shasum -a 256 -c SHA256SUMS.txt
xattr -dr com.apple.quarantine "/Applications/Fixer.app"

Known limitations

  • Gemini is the only provider in 0.2.0; transcription requires a network
    connection and the user's Gemini quota.
  • There is no local speech recognition or live partial transcript yet.
  • A {voice} Action normally performs one transcription request and one request
    to its configured text model.
  • When Fixer cannot prove that the original insertion target is unchanged, it
    deliberately leaves the result on the clipboard for manual paste.
  • The downloadable build is not notarized until the project has an Apple
    Developer ID release identity.

Roadmap

Planned directions, without promised dates:

  • more cloud API providers beyond Gemini;
  • optional local models for offline text processing and speech recognition;
  • a privacy-first Quick Insert library for reusable information such as
    addresses, phone numbers, signatures, and common text.

Quick Insert will contain only information the user explicitly saves. It is not
automatic history for selected text, Prompts, recordings, or transcripts. See
the full roadmap.

Assets 4
Loading