Free public beta · for Mac

TheSecretary

The oldest job in the office. Powered by the newest AI.

Speak naturally. The Secretary captures your words, lets you choose the engine, and turns speech into something useful. Hold a key while you talk-or press once to start and once to stop-then the completed thought appears where you were working.

6 speech enginesLocal or cloudVocabulary hintsHotkey testing

Version 1.0 beta (4) · Updated 2026-10-12 · macOS 13 or later. Apple silicon or Intel. Signed and notarized. Optional cloud services bill separately. Please help us test real-world dictation before the final release.

Start simply. Explore when you are ready.

A staff of one-working for you.

Start with Apple Speech or Local Whisper for transcription without per-minute provider fees. Cloud engines and paid text analysis are optional, not prerequisites.

Speak

Hold your hotkey, or toggle recording on and off. Paste the finished text into the app where you are working.

Words → accurate record

Choose

Compare local and cloud engines. Add vocabulary hints for your names, products and technical terms.

Choices → right service

Review

Find dictations in the Log. Optionally request summaries, action items, key points, chapters and sentiment.

Draft → stronger result

Enrichment is a separate Gemini or OpenAI text request, including when transcription is free. It appears in the Log when ready, not in the pasted dictation. Automated Concierge and Coach roles are future ideas, not features of this beta.

The visual cabinet

Five faces of the same capable helper

Typewriter, Dictaphone, shorthand notebook, Secretary and stenographer’s pad all belong to one retro office world. These illustrations give our new Mac app a familiar personality.

The Secretary identity board with five retro office icon directions and Playfair Display typography — Beta artwork · Playfair Display · cream paper · brass · dark green · oxblood. The Secretary records when you start it; it is not an always-listening assistant.

The modern relatives

One system mark, three backgrounds

The voice, six choices, and finished text stay the same. Only the atmosphere changes. These earlier studies can still represent the six-engine system underneath the retro personality.

Dark blueFocused · premium · strongest contrast

WhiteOpen · clean · approachable

Light blueFriendly · bright · conversational

One app. Six approaches.

Pick the strengths you need today

The fastest engine is not always the most private, the broadest in language support, or the best at cleaning up a wandering thought. The Secretary puts all six choices in one place-with room for more as speech technology improves.

Compare	Apple Speech	Local Whisper	Gemini 3.5 Transcribe	GPT Transcribe	GPT Live Transcribe	Meta Muse
Runs	On Mac	On Mac	Cloud	Cloud	Cloud stream	Cloud
Result timing	After stop; preview optional	After stop	After stop	After stop	Streams; final after stop	Experimental
Words while speaking	Local preview	No	No	No	Yes	No
Language reach	Varies by Mac	Multilingual; varies by model	Varies by provider	Varies by provider	Varies by provider	Unverified
Cleanup	Limited	Conservative local cleanup	Prompt guidance	Prompt guidance	Prompt guidance	Unverified
Vocabulary help	Context hints	Prompt hints	Prompt guidance	Prompt guidance	Live keywords	Keywords
Speaker labels	No	No	Not in app	No	No	Optional
Translation	Via Gemini	English locally; others via Gemini	Via Gemini	Via Gemini	Via Gemini	Via Gemini
Audio privacy	Local	Local	Google cloud	OpenAI cloud	OpenAI cloud	Meta cloud
API key	No	No	Gemini	OpenAI	OpenAI	Meta + billing
Provider charges	No per-minute fee	No per-minute fee	Your provider account	Your provider account	Your provider account	Experimental; verify billing

Recognition speed and quality vary with your Mac, microphone, language and provider. Meta access, coverage, billing and reliability remain experimental. Test important recordings before relying on any engine.

Choose objectively

Six engines, one clear comparison

The table describes what The Secretary exposes. A provider may offer other capabilities outside the app. Swipe or scroll sideways on a smaller screen.

Engine	Where and when	Best at	Cleanup	Vocabulary	Languages	Live behavior	Cost
Apple Speech	On this Mac Final result after stop; optional local partial words	Everyday speed, privacy and offline use	Apple punctuation; limited cleanup control	Up to 100 contextual hints	Varies by macOS locale and downloaded resources	Local preview	Free
Local Whisper	On this Mac Final result after release; Base or Small model	Private multilingual work without API fees	Model punctuation; conservative local cleanup	Private prompt guidance	Multilingual; quality varies by language and model	No provider partials	Free after 145 MB or 483 MB download
Gemini 3.5 Transcribe	Google cloud Completed recording uploads after release	Polished Smart dictation and code-switching	Strong punctuation, filler removal and paragraphs	Prompt guidance in this build	Provider-dependent; target translations use Gemini afterward	No provider partials in this build	Paid provider account; estimate shown in the app
GPT Transcribe	OpenAI cloud Completed recording uploads after release	Accuracy, accents, code-switching and technical names	Smart or Verbatim prompt guidance	Included in the transcription prompt	Broad multilingual recognition; target translations use Gemini	No provider partials	Paid provider account; estimate shown in the app
GPT Live Transcribe	OpenAI cloud Streams while you speak; saved-file fallback	Low-latency cloud words while talking	Smart or Verbatim guidance; early words may change	Live keyword hints plus fallback prompt	Multilingual live speech and code-switching	Genuine cloud live text	Paid provider account; estimate shown in the app
Meta Muse	Meta cloud · experimental Prepared WAV uploads after release	Testing keyword hints and optional speaker labels	Not verified enough for a dependable comparison	Keyword metadata	Coverage for this endpoint is not publicly verified	No provider partials; optional speakers	Experimental; verify access and billing before use

The app's cost figures are estimates, not invoices. Provider prices, taxes, free tiers and account terms can differ. Meta Muse access, coverage and pricing remain experimental.

Requirements

Ready on macOS 13+

A microphone and Microphone permission
Speech Recognition permission for Apple Speech
Accessibility permission for modifier-only hotkeys
Internet and your own API key for cloud engines
145 MB or 483 MB for an optional Local Whisper model

Vocabulary

One list, adapted to each engine

Add names, products and technical terms once. The Secretary sends them to the selected engine as contextual hints, prompt guidance or keywords. It does not alter the separate vocabularies used by Apple Dictation or Voice Control, and no engine can guarantee a hinted spelling.

Licensing

App and API charges stay separate

This public beta is free. Purchase reminders and evaluation-day counting are inactive until a future build has a working store. Saved API keys and preferences are not erased by reminders. Google, Meta and OpenAI accounts are always separate and bill their own API use. Apple Speech and Local Whisper have no per-minute provider fee.

Privacy follows the engine

Keep audio local, or choose cloud intelligence.

Apple Speech and Local Whisper can keep audio on the Mac. Gemini, Meta and OpenAI send audio to their services. Optional target-language translation sends the finished transcript text to Gemini, and AI enrichment sends finished text to Gemini or OpenAI. API keys are kept in Apple Keychain.