CLICK IT. TALK.Stops on its own when you pause.
Click the pearl and it melts into the orb. Stop talking for a second and a half and the take ends by itself. Prefer to click again? That is one toggle.
Free · On-device · No account
Private voice
dictation for Mac.
Hold a key and your words land at the cursor. Or click the buddy, read what you said, and drag it into any window. Speech recognition stays entirely on your machine.
Made by Billy Xu · Meet the developer ↗
Version 0.2.4 · 5.5 MB · macOS 13+ · Apple Silicon
Try the gesture
0:00
Ready · hold ⌥ to talk
Hold the key (mouse, finger, or the real Option key on your keyboard) and let go to drop the text at the cursor.
A simulation in your browser. Nothing is recorded; the real app runs on your Mac.
Signed & notarized by AppleNo accountNo subscriptionFree downloadAudio stays local
New in 0.2
Say it first.
Decide where it goes later.
Every voice-to-text tool drops your words wherever the cursor happens to be. Buddy mode holds what you said on screen instead. Read it, then drag it exactly where it belongs. Off by default; one switch in the menu bar.
CLICK IT. TALK.
DRAG IT ANYWHERE.Your words are a thing you can pick up.
The transcript lands in a glass bubble. Drag it into an editor, a chat box, a browser field, Terminal, even the Desktop. Nothing is typed for you unless you ask.
HOVER FOR THE REST.History, auto-stop, the menu.
Rest the mouse on the pearl and three buttons fan out. The last 50 takes are one hover away, kept in a file on your Mac and nowhere else.
SEE WHAT IT HEARD.The language mix, next to the text.
Whisper picks one language per take. When a mixed sentence comes out in one of them, the line under it, English 59% · 中文 39%, is the reason. It stops hiding the guess.
Hotkey takes are unchanged: hold the key and the text still types into the focused app. Watch the full demo and the setup steps on the Learn page.
One key
Or stay where you’re already typing.
The original gesture still works everywhere. There is no recorder window and no transcript to copy. PushSense works at the system level, then returns the text to the cursor you already placed.
- 01⌥⌘ ⌃ ⇧
Hold
Hold your push-to-talk key in any app. Right Option is the default; Right Command, Control, and Shift are selectable in the menu bar.
- 02
Speak
Whisper runs locally with Metal acceleration.
- 03
Release
The transcript is typed at your cursor.
Why PushSense
Your voice is input.
Not inventory.
PushSense keeps the short path from microphone to text on the Mac in front of you.
One input method, everywhere.
Use it in Mail, Notes, Slack, browsers, editors, and any other app with a text field. No plugins or app-specific integrations.
Nothing to upload.
The selected Whisper model runs on-device. After model setup, speech recognition works without an internet connection.
Speak how you think.
Automatic language detection means you never stop to pick a mode — including mid-sentence switches. Chinese and English are tested today, and more languages are being verified one at a time.
A real sentence
No mode switch,
mid-sentence.
“这个 feature 的 latency 有点高，我们 follow up 一下。”A real mixed-language sentence, transcribed in one pass.
Chinese + English verified · more languages in progressMeasured
Small app. Published numbers.
These are measurements from one Apple Silicon machine and one 27.7-second mixed Chinese/English recording—not universal performance promises.
- 5.5 MB
- Installed native app
- 181 MB
- Recommended Small q5_1 model
- 1.11 s
- Measured transcription time
- 25×
- About realtime in that test
Peak RAM was 484 MB with the recommended model in the same test. Read the benchmark and methodology ↗
First run
Download, grant, speak.
PushSense is signed with Developer ID and notarized by Apple. The built-in setup guides you through the permissions macOS requires. Watch the demo and read the quick start on the Learn page; every menu item, permission and fix is in the user guide.
- 1
Move it to Applications
Open the disk image and drag PushSense into Applications.
- 2
Grant three permissions
Microphone captures speech; Input Monitoring sees the global hotkey; Accessibility types the result into other apps.
- 3
Choose a local model
Download Small q5_1 (181 MB) for the current recommended balance of size and mixed-language accuracy.
FAQ
Before you install.
Is PushSense free?+
Yes. PushSense is free to download. There is no account, trial limit, or subscription.
Does PushSense upload my audio?+
No. Speech recognition runs on your Mac with whisper.cpp. The app downloads a model you choose, then transcription works offline.
Which Macs are supported?+
PushSense 0.2.4 requires macOS 13 or later and an Apple Silicon Mac.
Can I change the push-to-talk key?+
Yes. Right Option is the default, and the menu bar lets you switch to Right Command, Right Control, or Right Shift. PushSense uses a modifier you already have on the right side of the keyboard so it never collides with an app shortcut.
Why does it need Accessibility permission?+
Accessibility permission lets PushSense send the finished text as keystrokes to the field that already has focus. It does not replace your clipboard.
Which languages does PushSense support?+
The underlying Whisper models are multilingual and language detection is automatic, so you never pick a mode — including switching mid-sentence. Chinese and English are what I use daily and have tested; support for more languages is being verified and documented one at a time rather than claimed up front.
Which speech model should I use?+
Small q5_1 is the current recommendation. It is a 181 MB download and offered the best size-to-accuracy balance in the published mixed-language test.
What is Buddy mode?+
A small pearl that sits on your screen. Click it, talk, and the transcript waits in a glass bubble beside it instead of being typed anywhere. Drag the text into any window, or hover the bubble to copy or paste it. It is off by default; turn it on from the menu bar. The hotkey keeps working exactly as before.
Does dragging work in every app?+
Any window that accepts dropped text: editors, chat boxes, browser fields, Terminal, even the Desktop, where it becomes a text clipping. If a window refuses the drop, PushSense clicks where you released and pastes instead.
How long can I record?+
Each take is limited to five minutes in both hotkey and Buddy modes. At the limit, PushSense stops and transcribes automatically. Interrupted or unsaved takes remain as local WAV files under Recovered audio in the menu.
Where is the history kept?+
The last 50 takes live in a JSON file under Application Support on your Mac. Nothing is synced or uploaded. Hover the pearl or right-click it to open the list, and Clear history is at the bottom of every list.
What does "English 59% · 中文 39%" next to a transcript mean?+
Whisper decodes each take with one language. On a sentence that mixes two, it has to pick, and sometimes the whole sentence comes out in one of them. PushSense shows what the language detector heard so that when that happens, you know why.
What is Smart mode?+
An optional AI cleanup step for the transcript text. Use Local Ollama and nothing leaves your Mac, or your own OpenAI or Ollama Cloud API key. Audio never leaves your Mac either way. With Smart mode off, nothing is sent anywhere.
Do I have to say commands for Smart mode?+
No. Smart mode is a list of auto-fix rules: fix spelling, fix grammar and punctuation, remove filler words. Every enabled rule applies to everything you dictate. You can add your own one-line rules and edit the filler-word list per language.
Where is my API key stored?+
In the macOS Keychain, never in a plain file. Each provider keeps its own key and model, so you can set up OpenAI, Ollama Cloud, and Local Ollama and switch between them.