Press fn, talk, release. Your words paste into the active app.
Four model sizes: Small (466 MB), Medium (1.5 GB), Turbo (809 MB), Large-v3 (3.1 GB). GPU-accelerated on Apple Silicon via Metal.
Push-to-talk or toggle. Pastes into whatever has focus — Slack, Xcode, Mail, Figma, Cursor, the URL bar.
OpenAI, Anthropic, Groq, OpenRouter, Cerebras, Z.AI, or Apple Intelligence on-device. Only the transcript is sent — never audio.
One shortcut captures the screen, records the question, sends both to a vision model, pastes the answer.
Say "make it shorter" or "change Monday to Tuesday" and the previous output gets rewritten in place. No re-typing.
Different system prompt + vocabulary per active app. Terse for Slack, full sentences for Mail, code-aware for your editor.
Map spoken phrases to keystrokes. "approve" sends ⏎, "reject" sends ⎋. Useful with Cursor and Claude Code.
Say it mid-sentence; everything before drops. Configurable trigger phrase so you can self-correct without re-recording.
Whisper by default. Swap in Parakeet (Nvidia) for faster inference or Moonshine for English-only streaming. Same UI.
Different jobs, same shortcut. Ghostly adapts to the app in focus.
Dictate intent to Cursor and Claude Code at prompt speed. Map "approve" to ⏎ and "reject" to ⎋ so you never touch the keyboard.
Reply to Slack, Linear, and Jira without breaking focus. Per-app profiles keep Slack terse and email paragraphs clean — no re-editing.
Turn hallway thinking into sendable emails. Dictate the gist; AI cleanup shapes it into a paragraph you'd actually send — not a transcript.
First drafts at speaking pace. Cleanup fixes punctuation and fillers but keeps your voice — not a ChatGPT rewrite unless you ask for one.
Dump call notes into HubSpot or Salesforce while it's fresh. Between meetings, not at the end of the day when you've forgotten the details.
Brainstorm out loud straight into Figma comments, Notion, or your sketch app. Ideas land where you're working without a context-switch to a voice memo app.
Most people type around 40 WPM. Most people speak around 150. Ghostly narrows the gap between how fast you think and how fast your tools can keep up.
Most AI dictation tools ship your audio to a server and bill you monthly. Ghostly does neither — on-device transcription, choice of engine, and a one-time price.
Scroll horizontally to compare →
|
Ghostly
|
Wispr Flow | Willow Voice | Voicy | macOS Dictation | |
|---|---|---|---|---|---|
| Private by default — audio never sent to a server | Yes | No — cloud only | Opt-in mode | No — cloud only | Yes |
| Works offline | Yes | No | Opt-in mode | No | Yes |
| AI cleanup (punctuation, fillers, formatting) | Yes — BYO key | Yes — included | Yes — included | Auto punctuation | No |
| Choice of transcription engine | Whisper · Parakeet · Moonshine | Proprietary | Proprietary | Proprietary | Apple only |
| Configurable per-app prompts | Yes | Auto style | Auto format | Auto format | No |
| Screenshot + ask a vision model | Yes | No | No | No | No |
| Monthly subscription | Noneno subscription | $15/ month | $15/ month | $8.49/ month | Freebuilt-in |
| One-time / lifetime | $39one-time | —not offered | —not offered | $220lifetime | —n/a |
| Free tier | 30 min / wkall features | 2k words / wkcapped | 2k words / wkcapped | 40 min trialone-time | Unlimitedbuilt-in |
Whisper transcribes English, Spanish, French, German, Italian, Portuguese, Russian, Japanese, Mandarin, Arabic, Hindi — and plenty more — all on-device.
On-device. No cloud round-trip. No per-language setup.
The transcript pastes into whatever field the cursor is in. Silences get skipped automatically.
Any key combo. Push-to-talk or toggle. Voice-activity detection skips silences automatically.
Whisper transcribes on-device. Say "scratch that" to self-correct. Numbers get converted by AI cleanup.
Cleaned transcript pastes where your cursor is. No window switch, no copy step, no second thought.
Connect an API key for OpenAI, Anthropic, Groq, OpenRouter, Cerebras, Z.AI, or any OpenAI-compatible endpoint. Ghostly runs the transcript through a model before pasting — fixing punctuation, formatting lists, converting "twenty-five percent" to 25%, stripping fillers.
On Apple Silicon, pick Apple Intelligence for on-device cleanup with no key and no network.
Without a key, nothing breaks — you get the raw Whisper output.
um so I was thinking we could maybe push the launch to like the twenty fifth and then get marketing involved earlier this time uh yeah.
I was thinking we could push the launch to the 25th and get marketing involved earlier this time.
Ghostly adapts to whatever's in focus. Same voice input — different punctuation, casing, and tone per destination. Sound like you, not a transcript.
$ ghostly --toggle-transcription # toggle recording
$ ghostly --cancel # cancel current op
$ ghostly --start-hidden # launch to tray
$ ghostly --no-tray # quit on close
$ ghostly --debug # verbose loggingTrigger Ghostly from a Raycast script, a Stream Deck macro, or a shell alias. Remote-control flags send to the running instance — no duplicate process.
Ghostly pastes into whatever's in focus — editors, inboxes, chat, docs, terminals. Wherever your cursor lands.
Ghostly
Full feature parity on the free tier — every Whisper model, AI cleanup, all of it. Pro uncaps your time and lets you run it across three Macs. One-time purchase, no subscription.
Full feature parity, capped at casual-use volume.
Uncap the limits. Pay once, keep forever.
License key delivered instantly by email. Activates every Mac you own.
Ghostly is Apple Silicon only — any Mac with an M1 chip or later. Grab the latest .dmg, drag Ghostly to Applications, launch. It'll ask for microphone and accessibility access. Pick a shortcut in Settings. First run downloads a transcription model (500 MB–1.6 GB depending on which one you pick). After that, dictation works offline.
No. Transcription runs on-device via Whisper. If you enable optional AI cleanup with a cloud provider, only the resulting text is sent — never audio. Apple Intelligence keeps cleanup on-device too.
Yes. The free tier has full feature parity — all Whisper models, AI cleanup, per-app prompts, screenshot vision — capped at 30 minutes of transcription per week on one Mac. If you hit that cap, Ghostly Pro is a one-time $39 purchase that removes the time cap and lets you install on up to three Macs. No subscription.
macOS 12 or later on Apple Silicon — any Mac with an M1 chip or later. Intel Macs are not supported. Microphone and accessibility permissions required.
Ghostly runs every AI model on-device — Whisper transcription via Metal, and optional Apple Intelligence cleanup via the Neural Engine. Real-time local dictation needs the unified memory and ML accelerators that only Apple Silicon has. Supporting Intel would mean either unusable latency or shipping your audio to a server — a trade-off we won't make.
Yes. After the first-run model download, dictation works offline. Optional AI cleanup needs a network connection unless you pick Apple Intelligence.
No. Ghostly is Apple Silicon only — M1 or later. Intel Macs, Windows, and Linux are not supported.
Cursor
Claude
Windsurf
v0
GitHub
Xcode
VS Code
ChatGPT
LinkedIn
Linear
Figma
Notion
Google Docs
Discord
Terminal
Mail
Drafts
Obsidian
Outlook
Perplexity
Substack
Medium
Word
Canva
Instagram
YouTube
Things
WhatsApp
Zed
Notes
Superhuman
Salesforce
HubSpot
Replit
Calendar
AntiGravity