PushSense 0.2.4

About PushSense

• A new About PushSense item at the bottom of the menu opens a small window with the version and a quick link to send feedback.
• PushSense now carries its maker's mark: Billy Xu's signature, which goes on every app he makes. The menu item and the About window both show it, and it links to billyxu.dev.

Includes everything from 0.2.3. Requires Apple Silicon and macOS 13 or later. Signed with Developer ID and notarized by Apple.

PushSense 0.2.3

Clearer Buddy controls

• Hover over any of the three Buddy buttons to see what it does: recent transcripts, recording stop behavior, or the PushSense menu.
• The stop-mode description explains the current setting and what clicking changes. It updates immediately when you switch modes.
• Descriptions support English and Chinese, stay within the screen, and leave clicks and keyboard focus with your app.

Includes the Buddy drag fix from 0.2.2. Requires Apple Silicon and macOS 13 or later. Signed with Developer ID and notarized by Apple.

PushSense 0.2.2

Buddy Mode drag fix

• The pet now stays attached to the spot you grabbed instead of drifting away from the cursor during a drag.
• Hover controls stay hidden while you move the pet.
• Control-click opens history without also starting a recording when you release the mouse.

Requires Apple Silicon and macOS 13 or later. Signed with Developer ID and notarized by Apple.

PushSense 0.2.1

Safer everyday dictation

• Every transcription now has a 5-minute recording limit, in both hotkey and Buddy modes. At the limit, PushSense stops and transcribes automatically. The limit is visible in the menu.
• Recording buffers are bounded. If audio capture or saving cannot keep up, recording stops with a warning.
• Audio is checkpointed locally during recording. Interrupted takes and takes whose transcript could not be saved remain available through Recovered audio…. A new take does not overwrite them.
• Successful transcripts are saved to the existing local history in both hotkey and Buddy modes. History still keeps the latest 50 entries.

Local storage

Recovery WAVs are stored in ~/Library/Application Support/PushSense/DictationRecovery/. They are removed after a successful transcript is saved to history. Interrupted/unsaved takes remain until you remove them. The recovery folder has a 200 MiB budget; reaching it prevents new recordings instead of deleting older takes. Audio is never uploaded. Recovery cannot guarantee against power loss or disk failure.

Existing workflows

Keep using hold-to-talk, Buddy click/stop, drag/copy/paste, language selection, and optional Smart mode as before. This release does not add meeting notes or long-form recording.

The native transcription libraries are rebuilt for the advertised macOS 13 deployment target using the same upstream whisper.cpp and ggml versions. Requires Apple Silicon. Signed with Developer ID and notarized by Apple.

PushSense 0.2.0

Buddy mode: click a pearl to talk, then drag the words anywhere

Until now PushSense had one gesture: hold a key, speak, and the text lands where
your cursor is. That order is fine when you know where the words go. It is
awkward when you want to say something first and decide later.

Buddy mode puts a small pearl on your screen. Click it and it melts into
the orb you know from the halo; a capsule beside it shows a red dot, a timer
and your voice. Stop talking for a second and a half and it stops by itself
(or switch that off and click again). The transcript then lands in a glass
bubble next to the orb.

The text in that bubble is a thing you can pick up. Drag it into any
window that takes text (an editor, a chat box, a browser field, Terminal,
even the Desktop, where it becomes a text clipping) and let go: the words land
at the drop point. Nothing is typed for you unless you ask. If a window refuses
the drop, PushSense clicks where you released and pastes instead.

Also in the bubble, on hover: copy, paste into the focused app, history, and
close. After 25 seconds of being ignored the bubble folds into a one-line
pill; hover to unfold it.

Buddy mode is off by default: turn it on from the menu bar. The hotkey
works exactly as before whether the pearl is on screen or not, and a hotkey
take still types into the focused app. By default a buddy take stays in the
bubble; Buddy output can make it paste like a hotkey take.

Hover the pearl for the small things

Rest the mouse on the pearl and three glass buttons fan out over it:

• History, the last 50 takes. Pick one and it is back in the bubble,
  ready to drag or paste. Right-clicking the pearl opens the same list, and
  the menu bar has it too.
• Auto-stop, the ear: on, a take ends after 1.5 s of silence; off, it
  ends when you click the pearl again. The capsule says which.
• Menu, the menu bar dropdown, without reaching for the menu bar.

More will live here later (the speech model and language switches are the
obvious candidates). History is a JSON file in Application Support, on your
Mac only, and Clear history is in every history list.

See what the language detector heard

Whisper decodes a take with one language token. On a sentence that mixes
English and Chinese it has to pick, and sometimes the whole sentence comes out
in one of them. PushSense already leans Chinese when Chinese is plausibly
present; it will never be right every time.

So now it tells you. The bubble's meta line and every history row carry what
the detector heard, English 59% · 中文 39%, leader first. When a mixed
sentence comes out in one language, that line is the reason.

Smaller things

• A resting pearl is a still image, not an animation: buddy mode costs nothing
  while you are not using it, and the cursor halo is only created when it is
  actually needed.
• The bubble, capsule and buttons are Liquid Glass on macOS 26; earlier
  versions get a frosted panel with the same layout.
• Buddy strings are in English and Chinese, like the rest of the app.

---

Signed with Developer ID and notarized by Apple: double-click to open, no
right-click workaround needed. Requires macOS 13 or later, Apple Silicon.

PushSense 0.1.0

Updating no longer sends you to a web page

Check for Updates used to find the new version and then hand you to a browser:
you downloaded it, found it in Downloads, and opened it yourself. The app that
told you about the update did none of the work.

Download now fetches the .dmg itself, with a progress window, and opens it
when it lands. The install is the drag and nothing else.

• The file goes to your real Downloads folder, resolved through the system
  rather than a hand-built path, so a relocated or localized folder still works.
  If it cannot be written there, a temporary copy still installs.
• A failed request is caught before the response is treated as a disk image.
  Without that check, an error page would arrive renamed .dmg.
• A release with no .dmg falls back to opening the release page.

The drag stays yours: PushSense is running from the bundle being replaced, so
it cannot overwrite itself while open.

Tell me what's wrong, in one sentence

New in the menubar: Send Feedback…. It opens a short form with your version,
macOS version, speech model and Smart mode provider already filled in, so a
report like "this word never comes out right" arrives with the context that
makes it fixable.

The app sends nothing on its own. It opens the page, you see every value before
you send, and the email field is optional and used only to reply. Details are in
the privacy policy.

Bug reports and feature requests are still welcome in the
issue tracker if you
prefer a public thread.

---

Signed with Developer ID and notarized by Apple: double-click to open, no
right-click workaround needed.

PushSense 0.0.9

Smart mode now runs fully on your Mac

Smart mode could only talk to a cloud provider — dictation had to leave the Mac
to get cleaned up, and needed a paid key to do it. Local Ollama now joins
OpenAI and Ollama Cloud as a peer provider: no key, no account, and the
transcript never leaves the machine.

• Pick the model from what Ollama actually has installed, rather than typing a
  name (a typo there is a 404 you cannot diagnose).
• "Ollama isn't running" and "no models pulled" are now worded differently —
  both are actionable, and modelNotInstalled names the exact ollama pull to run.
• One resident model: it loads when Smart mode uses it and unloads otherwise, so
  an unused feature costs no RAM. keep_alive is re-asserted on wake.

Also in this release

• The filler rule's language chips are shown expanded — hidden behind Edit, the
  rule looked like it did nothing when the presets did not match what you spoke.
• Save rules now confirms the write (rules always saved on change, which was
  invisible and read as broken).

---

Unsigned build. macOS will block it on first launch — right-click → Open
once, as the included READ ME FIRST.txt explains.

PushSense 0.0.8

Small fix

• Filler-word presets are now English and 中文 only. The Spanish and Japanese lists implied dictation support the app does not have yet; your own words still work for any language.

Everything from 0.0.7 (中文界面, Smart mode with your own OpenAI / Ollama Cloud key) is unchanged. User guide · 中文说明

PushSense 0.0.7

中文界面 · App language

PushSense now speaks Chinese. App language in the menu bar switches every menu, window, and message between English and 中文; System follows your macOS language. The tutorial and the Smart mode window re-render in place when you switch.

菜单栏新增「界面语言」：中文 / English / 跟随系统。所有菜单、窗口、提示都会翻译，切换立即生效。

Smart mode polish

• Provider tabs (OpenAI | Ollama Cloud) are always visible, so adding a second provider never touches the first one's key.
• Filler-word rule has per-language presets (English, 中文, Español, 日本語) plus your own words.
• Saved keys show masked with the last four characters.

Full details: user guide · Learn page · 中文

PushSense 0.0.6

Smart mode: bring your own key

Optional AI cleanup of your dictation, with your own API key. Dictation itself stays fully local; only the transcript text goes to the provider you pick.

• OpenAI or Ollama Cloud. Paste a key, press Test & save. Each provider keeps its own key (macOS Keychain) and model, so you can set up both and switch.
• Auto-fix rules, no spoken commands. Fix spelling, fix grammar and punctuation, remove filler words: every enabled rule applies to everything you say. Add your own one-line rules ("Write numbers as digits").
• Filler words per language. English, 中文, Español, 日本語 presets plus your own words.
• Raw mode is unchanged: with Smart mode off, nothing is sent anywhere.

Every AI request logs model, status, and token count (never the text or key) to the runtime log, so you can confirm what left your Mac.

Full details: user guide.