OpenDMG
Speech Wingman icon

Speech Wingman for Mac

Fully offline speech companion that nudges you when your words match rules you set.

By AmethystineAlpaca. Free and open source, MIT licence.

Get it with OpenDMG View source on GitHub

Speech Wingman at a glance

Version
0.4.0 (pre-release)
Released
30 September 2026
Download size
299 KB
Works on
Apple silicon and Intel
Licence
MIT
Category
Audio and Music
Developer
AmethystineAlpaca
GitHub stars
1

About Speech Wingman

Private, offline AI speech companion for macOS. Multilingual: English, Chinese, Japanese, Korean, and Cantonese, including language switching. Write your own reminder rules. Runs locally on Apple Silicon.

Screenshots

Speech Wingman screenshot 1Speech Wingman screenshot 2Speech Wingman screenshot 3

How to install Speech Wingman on a Mac

  1. Get OpenDMG. It is a free app store for open source Mac apps. Download OpenDMG.
  2. Search for Speech Wingman and press Get. OpenDMG downloads version 0.4.0 from AmethystineAlpaca's own GitHub release.
  3. Open it. The download is checked against its SHA-256 fingerprint and its signature is checked on your Mac before it lands in Applications.

Updates to Speech Wingman show up in OpenDMG and install with one button. No account is needed.

Prefer to do it by hand? The developer's file is on GitHub: Speech-Wingman-Ultimate-Evaluation-v0.4.0.zip (299 KB).

What is new in Speech Wingman 0.4.0

Speech Wingman 0.4.0 brings live transcription to the floating control, improves continuous-speech evaluation, and calibrates the actual sensitivity mapping against measured recall.

## What changed

- **UI:** recent and live transcript, automatic follow, retained text when paused, native dragging, and a settings shortcut. Settings now include a larger rule editor, character count, expandable examples, sensitivity cards, a pinned save footer, save feedback, and Command-S. Updated English/Chinese layouts and screenshots.
- **Continuous speech:** bounded statement grouping, four-statement FIFO, explicit one-step deferred continuation, and a 30-second freshness limit. New speech no longer cancels a running judgment.
- **Global inference harness:** route large rule lists before independent checks, separate rule instructions from speech, validate evidence, constrain generated output, and bound malformed-output repair. No rule-topic-specific patches.
- **Sensitivity:** for 8+ rules, current Low/Medium/High selects legacy Low/High/Medium. Smaller rule lists retain their original mapping. Saved settings keep their identity; exports record `routed-recall-v1`.
- **Diagnostics:** exported configuration snapshots, inference inputs, segment IDs, matched rules, continuation links, validation failures, timings, queue status, and reminder disposition.
…

More Audio and Music apps

All Audio and Music apps