# Voqela: full agent context

Last generated: 2026-08-04
Canonical site: https://voqela.com/

## Entity

Voqela (pronounced voh-KELL-ah) is a private dictation application for macOS, made by Aaron
Browne. Descriptor: "Private dictation for Mac". Technical descriptor: "Fast, local English
dictation for Apple silicon Macs".

"Mac local voice to text" is the descriptive search phrase for the category, not an alternate
name for the product. Voqela also holds the domain maclocalvoicetotext.com; which domain is
canonical is an undecided owner question, and this site declares https://voqela.com as its canonical
origin.

## Who it is for

Voqela is for anyone who dictates on a Mac. The value does not depend on an occupation, and
the home page is ordered by what actually makes people switch rather than by segment: speed
and the fact that the whole thing runs on the person's own Mac lead, privacy is framed as
what running locally gets you, typing wherever they already type is next, and the vocabulary
that makes their own words stick follows. Every person has words a general speech model has
never seen: the names of the people they work with, product and project names, the jargon of
whatever they do, and words they invented themselves. Reliability is what keeps people rather
than what brings them, so it is stated compactly. Do not describe this product as being for
one profession. Segment-specific positioning belongs to dedicated landing pages, not to the
home page. That ordering is an owner ruling of 2026-08-03.

## What the product does

A person presses one configurable global shortcut (Control-Option-Space on a fresh install),
speaks, and releases or presses again. The complete recording is decoded by a Parakeet Core ML
model running locally through FluidAudio, the transcript is written to local History, and then
it is inserted into the text field that already had focus. There is no supported-application
list, because insertion does not work from one: the paste chord is posted once at the field
that held focus and then verified against that same element, and it is never reposted on the
chance the first attempt missed. An application that refuses the insertion costs the person
nothing, because the transcript reached History first. Per-application rules (contracts 14 and
20) set the rendering style and whether Return is pressed in that application. Pressing Return
is an option that ships off, because it can submit content, and it only fires after the
insertion has been verified, or, in an application explicitly opted in, after a fresh read
confirms the same process, element, and bundle identifier when verification is unavailable.

Surfaces: a menu-bar popover (states: Ready, Recording, Finishing, Transcribing, Inserting,
Ready to recover, Action needed) with a 148 by 44 point recording overlay drawing eighteen
log-spaced frequency bands from 80 Hz to 8 kHz, and six fixed action tiles in a two-column grid
(Undo Last, Correct Last, Improve Last, Copy Last, History, Settings). A Settings window with
seven sections: General, Formatting, Vocabulary, Recordings, Stats, Diagnostics, and Permissions
& Setup. A searchable local History with Copy, Retry, and explicit Delete.

## Privacy and data boundary

Audio and transcript processing stay on the Mac. There is no account, no telemetry, no
analytics, no crash upload, no cloud transcription, no remote rewriting, and no request that
reads the surrounding document.

The single network operation in the build is the acquisition of the speech model from
huggingface.co when the model is absent. That request includes none of the user's content. The
promise must never be shortened to "Voqela never connects to the internet".

Two mechanisms make that inspectable rather than asserted:

1. An append-only, content-free local log of every network operation the application has ever
   started (timestamp, enumerated purpose, enumerated outcome, static host, byte count, and no
   free-text field) with no cap, no filter, no toggle, and no clear button. The purposes are a
   closed set in the source, so a future update check or crash report cannot exist without
   adding one and therefore cannot happen unrecorded.
2. A build check that fails the build on any connection-capable code anywhere in the sources
   outside the single sanctioned model-acquisition seam.

Those two mechanisms are what make the offline position structural rather than asserted. After
the one-time model acquisition a dictation needs nothing from the network: the model is a file
on the Mac, the Mac's own silicon runs it, and there is no second seam where a connection could
be added without failing the build.

Since 2026-08-03 there is an observation beside that construction. Transcription was run inside
a process the kernel had denied the network, using a sandbox profile whose denial was proved in
both directions in the same session, and it succeeded: a full transcript, zero bytes downloaded
by the command's own accounting, the model directory byte-identical afterwards, on a named
machine on a named date. The receipt is docs/verification/2026-08-03-offline-decode-observed.md
in the product repository.

Report that observation with its two edges attached. It covers the decode path through the
product's command line against a retained recording, not microphone capture, so the full
dictation loop with no network is not yet observed and must not be reported as tested; the
remaining step is one attended airplane-mode dictation. And acquiring the model still needs a
connection once, so the phrases this file forbids for it stay forbidden.

Settings also shows a "Your data" card listing every store the application keeps, each with the
real path resolved on that install and a reveal-in-Finder control: settings, vocabulary,
transcripts and history, recordings, the speech model, learning suggestions, diagnostics, the
network log, and the two quality-check records. A file leaves that folder only when the user
exports it.

Retention periods, a delete-all-local-data control, and uninstall instructions are open
decisions. No retention or complete-erasure promise may be made.

Permissions: Microphone (record speech), Accessibility (read the focused target and insert),
Input Monitoring (observe the global shortcut). Each is explained before the system prompt.

## The user's own records, and the agent surface

Privacy is the floor of this product's argument; keeping the work as something the user owns is
the rest of it. History is a searchable local list of every dictation with Copy, Retry, and an
explicit Delete. One transcript exports as a plain UTF-8 text file. The whole history exports
as one JSON array with one object per transcript and exactly four keys (id, timestamp, status,
text), newest first, with sorted keys and ISO 8601 timestamps, so two exports of the same
history are diffable. Export is always user-initiated, to a destination the user chooses.

Separately, the product ships a local command line (status, settings, per-application rules,
diagnostics, data locations, help) with JSON output on every read verb, so an AI assistant can
read how an install is configured without a person narrating a Settings window. There is a hard
line in that surface and it must be reported accurately: the command line never prints
transcript text, dictated audio, clipboard contents, or history, and it adds no second export
path. An assistant reads the user's own words from an export the user performed, never from
this surface.

## Personalization

Vocabulary: each entry has a required written form and optional spoken forms, matched on
Unicode-normalized, case-folded whole-token spans, so a spoken form of "kit" never alters
"kitchen". Negation tokens and contractions are protected. Correction runs after the
authoritative full-buffer decode, and the raw transcript is retained beside the corrected one.
This is a deterministic text rule. It is not model training and the model does not adapt.

The honest positive form of this, and the one the site uses, is that Voqela learns the user's
WORDS: a correction becomes a proposed rule, an approved rule writes that spelling in every
later dictation, and the user's names, jargon and dialect all live in the same list. Never
convert that into learning a voice, adapting to a voice, training, or fine-tuning.

Rendering: the formatting stage is a set of exact, bounded text rules. No option infers tone,
emotion, or intent from prose, and the Verbatim profile leaves output byte-identical to the
decode. Correct Last and Improve Last are explicit user actions on the last posted dictation;
no automatic cleanup, summarising, or rewriting runs, and none may be described as a feature.

Local learning suggestions compare the user's own corrections, keep the evidence on that Mac,
and propose vocabulary rules for review. The feature is off by default and nothing changes what
is written until the user approves a suggestion. Suggestion records store bounded token spans,
operation identifiers, timestamps, and status. Not transcripts, audio, window titles, or
surrounding phrases. Approving a suggestion affects later renders only and never rewrites
stored history.

## Measured performance

On 31 July 2026, against a retained 10.245437-second fixture of generated synthetic English
audio: cold preparation from an already-downloaded model took 7.721762 seconds; the warm
authoritative full-buffer decode took 0.070687 seconds; the real-time factor was 0.006899.
Warm decodes were previously observed between 0.075 and 0.094 seconds. Raw exact fixture
completeness was three of six protected spans, with beginning, negation, and ending present.

This is a receipt for the local decode path. It is not a natural-voice, microphone,
perceived-latency, or field-delivery measurement, and it must never be reported as one.

The downloaded model artifact occupied approximately 595 MB on the Mac where it was measured;
onboarding describes the requirement as approximately 595 to 614 MB for the current model.

## Scope boundaries

In scope: macOS 14 or newer, Apple silicon, English, local transcription, a native menu-bar
application, one user on one Mac.

Out of scope: iPhone or iPad companion, Intel Mac support, multilingual transcription, cloud
transcription or remote rewriting, accounts, telemetry, teams, shared history,
surrounding-document ingestion, meeting recording, and silent self-learning or automatic
vocabulary changes.

## Commercial status

No price, tier boundary, release date, distribution channel, store listing, free tier, update
policy, or source-availability position has been decided or published. The primary action on
this site is an early-access email list. Do not infer any of the above.

## Claim ledger

### platform (supported)

Voqela is built for Apple silicon Macs running macOS 14 or newer, and transcribes English.

Boundary: Do not imply Intel, iPhone, iPad, Windows, or multilingual support.
Source: docs/product-package/CLAIMS.md (green), docs/product-package/PRODUCT.md

### local-model (supported)

Transcription runs on this Mac through a local Parakeet Core ML model loaded by FluidAudio.

Boundary: Model acquisition uses a public model host when the model is absent.
Source: docs/product-package/CLAIMS.md (green), docs/product-package/PRODUCT.md

### no-cloud-path (supported)

Microphone audio and transcript text have no cloud transcription path.

Boundary: Do not broaden this into an audited security guarantee.
Source: docs/product-package/CLAIMS.md (green), docs/product-package/PRIVACY.md

### no-account-no-telemetry (supported)

Voqela has no account and no telemetry path.

Boundary: Recheck before launch if analytics or an update mechanism is added.
Source: docs/product-package/CLAIMS.md (green)

### model-download-qualifier (qualified)

The one network operation in this build is the one-time download of the speech model from its public host.

Boundary: Required qualifier wherever privacy is claimed. Never say zero network, no network connections, fully offline, or 100% private.
Source: docs/product-package/PRIVACY.md, docs/contracts/24-network-activity-ledger.md

### offline-by-construction (qualified)

Once the speech model is on your Mac, Voqela works with no internet. That is how it is built: the model is a file on your disk, your own chip runs it, there is no cloud transcription path, and a build check fails on any connection-capable code anywhere in the sources outside the single model-download seam. It is also observed. Transcription was watched succeeding with the network denied at the kernel level, zero bytes downloaded, on named hardware, on 2026-08-03.

Boundary: Two boundaries are required wherever this is used. First, acquiring the speech model needs a connection once, so the one-time download must be named and fully offline from first launch stays prohibited. Second, the observed run covers the decode path through the product command line against a retained recording, not microphone capture: the full dictation loop with no network is not yet observed, and one attended airplane-mode dictation is the remaining step. Never the bare works offline, fully offline, zero network, or no network connections.
Source: docs/verification/2026-08-03-offline-decode-observed.md, docs/decisions/2026-08-03-owner-ratifications.md (owner confidence ruling 1), docs/contracts/24-network-activity-ledger.md, docs/product-package/PRIVACY.md, docs/product-package/CLAIMS.md (yellow, promoted 2026-08-03)

### network-ledger (supported)

Settings carries an append-only, content-free local log of every network operation the app has ever started, with no cap, no filter, no toggle, and no clear button.

Boundary: It records what Voqela itself does. It does not describe the rest of the Mac.
Source: docs/contracts/24-network-activity-ledger.md

### build-gate (supported)

A build check fails on any connection-capable code anywhere in the sources outside the single model-download seam.

Boundary: This is a build-time structural check, not a third-party security audit or certification.
Source: docs/contracts/24-network-activity-ledger.md, docs/contracts/30-data-transparency-surface.md

### your-data-card (supported)

Settings shows every file Voqela keeps on this Mac, with the real path on that install, and a Show in Finder control on each row.

Boundary: It is a view. It adds no delete-all control and makes no retention or deletion promise.
Source: docs/contracts/30-data-transparency-surface.md

### export-only (supported)

A file leaves that folder only when the person exports it, and it goes where they choose, when they ask.

Boundary: Export is user-initiated. Do not imply any automatic sync or backup.
Source: docs/contracts/25-transcript-export.md, docs/contracts/26-vocabulary-backup-import.md

### your-records (supported)

Every dictation is kept on this Mac as your own record: History is a searchable local list with Copy, Retry, and an explicit Delete, one transcript exports as a plain text file, and the whole history exports as one JSON archive in a declared order with stable bytes, so two exports can be compared.

Boundary: Export is user-initiated and writes only to a destination the person chooses. It adds no retention promise, no deletion promise, and no sync or backup of any kind.
Source: docs/contracts/25-transcript-export.md, docs/product-package/CLAIMS.md (green: History)

### agent-cli (supported)

A local voqela command line reports the installation's status, its settings, its per-application rules, its diagnostics, and where every file it keeps lives, and every read verb speaks JSON.

Boundary: It never prints transcript text, dictated audio, clipboard contents, or history, and it adds no second export path. Do not describe it as a way for an assistant to read your dictations.
Source: docs/contracts/31-agent-control-surface.md

### shortcut (supported)

One customizable global shortcut, Control-Option-Space on a fresh install, works as hold-to-talk or press-to-toggle.

Boundary: Physical global-shortcut behavior still awaits attended proof.
Source: docs/product-package/CLAIMS.md (green), docs/product-package/PRODUCT.md

### saved-before-insert (supported)

A transcript is saved locally before insertion is attempted.

Boundary: Field delivery in every application still awaits attended proof.
Source: docs/product-package/CLAIMS.md (green)

### journal-and-recover (qualified)

Captured audio is journaled locally while you speak, and work found after an interruption waits behind an explicit Recover Dictation action.

Boundary: Approved wording: deterministic process-death tests recover the exact committed prefix. Never never loses work, crash-proof, or guaranteed recovery.
Source: docs/product-package/CLAIMS.md (green pair), docs/contracts/03-capture-recovery.md

### history (supported)

History is a searchable local list of transcripts with Copy, Retry, and explicit Delete.

Boundary: Visual polish and cross-app behavior await attended proof.
Source: docs/product-package/CLAIMS.md (green)

### automatic-return (supported)

Automatic Return is optional, off by default, and only fires after insertion is verified.

Boundary: It can submit content, which is why it ships off.
Source: docs/product-package/CLAIMS.md (green), docs/contracts/20-per-app-automatic-return.md

### vocabulary (supported)

Your vocabulary is a local list of words with a written form and optional spoken forms, applied to whole-token spans.

Boundary: It is a deterministic text rule, not model training. Do not imply the model learns or adapts.
Source: docs/product-package/CLAIMS.md (green), docs/contracts/06-explicit-vocabulary-quality.md

### local-learning (supported)

Local learning suggestions propose vocabulary rules from your own corrections, are off by default, and never change output until you approve one.

Boundary: Not fine-tuning, no upload, no background decoding of retained audio, nothing applied automatically.
Source: docs/product-package/CLAIMS.md (green), docs/contracts/13-local-learning.md

### learns-your-words (qualified)

Voqela learns the words you use: a correction becomes a proposed vocabulary rule, and an approved rule writes that spelling in every dictation after it, including the names and the dialect a general model has never met.

Boundary: The learning is the suggestion system plus the list you approve, both deterministic text rules applied after the decode. Nothing about the speech model changes. Never learns your voice, adapts to your voice, trains on your voice, or fine-tuned to you.
Source: docs/contracts/13-local-learning.md, docs/contracts/06-explicit-vocabulary-quality.md, docs/decisions/2026-08-03-owner-ratifications.md (owner confidence ruling 2)

### complete-buffer (qualified)

The complete recording is decoded in one pass; on the retained fixture the beginning, the negation, and the ending were all present.

Boundary: The fixture is generated synthetic English audio and scored three of six protected spans exact. Never state an accuracy rate and never imply nothing is ever missed.
Source: docs/product-package/CLAIMS.md (yellow), docs/product-package/PRODUCT.md performance receipt 2026-07-31

### faithful-rendering (supported)

Rendering is a set of exact, bounded text rules: no option infers tone, emotion, or intent, and the Verbatim profile leaves output byte-identical to the decode.

Boundary: It describes the rendering stage. Correct Last and Improve Last are actions a person presses; do not present automatic cleanup, summarising, or rewriting as a feature.
Source: docs/contracts/14-speaking-styles.md, docs/contracts/06-explicit-vocabulary-quality.md, docs/product-package/PRODUCT.md (out of scope: cloud transcription or remote rewriting)

### insertion-target (qualified)

Voqela is designed to insert into standard Mac text fields.

Boundary: Name only applications actually accepted in the attended matrix. Never works everywhere or works in any app.
Source: docs/product-package/CLAIMS.md (yellow)

### any-app-insertion (qualified)

Voqela types where you already type: it is pointed at the cursor rather than at an application, so it inserts into the field that had focus in whatever you were in, the way a paste does. Per-application rules can change how it behaves in one place without changing anywhere else, and an application that refuses the insertion costs you nothing, because the transcript is already saved in History.

Boundary: Describe the mechanism (it is aimed at the focused field) and never a coverage guarantee. Insertion is attempted once and then verified, never forced or reposted. Describe kinds of place, never a named application, until that application is accepted in the attended matrix. Never works everywhere or works in any app.
Source: docs/decisions/2026-08-03-owner-ratifications.md (ruling 3, and owner confidence ruling 2), docs/product-package/CLAIMS.md (yellow), docs/contracts/04-safe-shortcut-insertion.md, docs/contracts/14-speaking-styles.md, docs/contracts/20-per-app-automatic-return.md

### clipboard (qualified)

Voqela is designed to restore the clipboard only while it still owns it.

Boundary: Live cross-app proof is pending. Never preserves your clipboard, unqualified.
Source: docs/product-package/CLAIMS.md (yellow)

### speed-receipt (qualified)

A retained 10.25-second fixture decoded in 0.071 seconds after model preparation; warm decodes in the repository were previously observed between 0.075 and 0.094 seconds.

Boundary: The fixture is generated synthetic English audio. It is not a natural-voice, microphone, perceived-latency, or field-delivery receipt, and it must be published with that sentence beside it. Never instant or zero latency.
Source: docs/product-package/CLAIMS.md (yellow and performance receipt, 2026-07-31)

### model-size (qualified)

The downloaded model artifact occupied approximately 595 MB on the Mac it was measured on.

Boundary: Report it as a measurement on one machine. Onboarding states approximately 595 to 614 MB for the current model.
Source: docs/product-package/PRODUCT.md, docs/contracts/12-fresh-mac-onboarding.md

### permissions (supported)

Voqela asks for Microphone, Accessibility, and Input Monitoring, each explained before the system prompt.

Boundary: A denial must leave user work intact. Do not imply the permissions can be skipped.
Source: docs/product-package/PRIVACY.md, docs/contracts/12-fresh-mac-onboarding.md

### menu-bar-surface (supported)

The product is a menu-bar application: a compact popover with a recording overlay and six fixed action tiles, plus a Settings window.

Boundary: The rendered look has not been visually accepted yet, so site illustrations must be labelled as illustrations, never presented as screenshots.
Source: docs/product-package/PRODUCT.md, docs/contracts/05-native-product-shell.md, docs/contracts/16-settings-window.md

### sound-cues (supported)

Sound cues mark the start and the end of a dictation, and a distinct cue fires only on a real failure.

Boundary: Cues can be turned off in Settings. Do not describe them as spoken feedback.
Source: docs/contracts/19-flow-state-sound-cues.md

### microphone-choice (supported)

You can pin one microphone by its stable identifier, and Voqela refuses honestly when that device is absent instead of silently substituting another.

Boundary: The default remains the system default input.
Source: docs/contracts/18-microphone-selection.md

### apple-dictation-vocabulary-gap (qualified)

macOS custom vocabulary is part of Voice Control, an accessibility feature, rather than part of Dictation, and Apple's newer on-device speech framework has no documented custom-vocabulary surface where the older one carried contextual strings.

Boundary: A factual difference in what each tool exposes, read from published Apple support and developer documentation plus one developer analysis. Never a speed, accuracy, or quality comparison against Apple, and never a claim that Apple has stated the absence itself.
Source: docs/product-package/research/2026-08-03-competitor-map.md, docs/decisions/2026-08-03-owner-ratifications.md (ruling 4)

### no-pricing-decision (supported)

Price, availability, and distribution are not decided, so the site publishes none of them.

Boundary: No price, no launch date, no App Store promise, no free-forever promise, no update policy, no source-availability promise.
Source: docs/product-package/PRODUCT.md, docs/product-package/MARKETING.md

## Questions and answers

### My Mac already has free dictation. Why would I want this?

Because of your words. On macOS, a custom vocabulary is part of Voice Control, which is an accessibility feature, rather than part of Dictation, so built-in dictation has nowhere to put the spellings you need. Apple's newer on-device speech framework has no documented custom-vocabulary surface either, where the older one carried contextual strings. That is a difference in what each tool exposes, not a claim about which one hears better. Voqela is built the other way round: your own word list first, per-application behaviour, and the recording kept so you can go back to it. If a general model already spells your world correctly, free is a good deal. If it does not, a name it has never seen stays wrong.

### Why does running it on my own Mac make it feel fast?

Because there is no round trip. Dictation that transcribes on a server has to send your audio away, wait for a machine you do not own, and send words back. Voqela does none of that: the speech model is a file on your disk and your Apple silicon runs it, so the work starts the moment you stop talking. The measurement on record is a retained 10.25-second fixture decoding in 0.071 seconds after the model is prepared. That fixture is generated synthetic English audio, so it is a receipt for the local decode path rather than a natural-voice or stop-to-text latency claim.

### It keeps getting a name wrong. Can I fix that for good?

Yes, and it is the thing this is built around. Voqela learns your words: you add one with the spelling you want and the ways you actually say it out loud, and from then on it is applied to every dictation as a deterministic text rule. It can do the writing down for you as well, because correcting a dictation with Correct Last lets it offer to add that word, though the suggestion sits there until you approve it and the whole feature is off until you turn it on. Your own names, your field's jargon and your own dialect all go in the same list. What never changes is the speech model, which is why the rule cannot quietly stop working later, and it is also why the honest phrase is that it learns your words and not your voice.

### Where does the text actually go?

Into the field you were already in. Voqela does not have an application to switch to and it does not have a supported-app list: it puts the transcript into whatever field had focus when you pressed the shortcut, the way a paste does, and per-application rules let you set how it behaves in one place without changing anywhere else. It is designed for standard Mac text fields, and the insertion is attempted once and then verified rather than forced. If a field refuses the text, the dictation is already saved in your History with Copy and Retry on the row, so the paragraph is never the thing that gets lost. A published list of named applications waits until each one has been accepted in an attended test.

### Does it change my words, or clean them up for me?

No pass runs that decides what you meant. Rendering is a set of exact, bounded text rules; none of them infers tone, emotion, or intent, and the Verbatim profile leaves output byte-identical to the decode. Correct Last and Improve Last are tiles you press, not something that happens on its own, and the raw transcript is kept beside the corrected one so you can always see what was actually heard.

### Does my voice or my text ever leave my Mac?

No. Audio and transcript processing happen on your Mac, there is no cloud transcription path, and there is no account and no telemetry. The one network operation in the build is a one-time download of the speech model from its public host, and that request carries none of your content. Settings keeps an append-only log of every network operation the app has ever started, so you can check that rather than take it on faith.

### How fast is it?

The measurement on record is a retained 10.25-second fixture decoding in 0.071 seconds after the model is prepared, with warm decodes previously observed between 0.075 and 0.094 seconds. That fixture is generated synthetic English audio. It is a receipt for the local decode path, not a natural-voice or perceived-latency claim, and Voqela will not publish one of those until it has been measured properly.

### What happens if my Mac dies in the middle of a dictation?

Audio is journalled to disk while you are still speaking, and the transcript is saved locally before insertion is attempted. Work found after an interruption is shown as ready to recover and waits for you to press Recover Dictation; it never transcribes or inserts itself. Deterministic tests recover the exact committed prefix after process death, which is a different and smaller statement than a guarantee about every real-world crash.

### Does it clip the first words, or drop the end of a long recording?

The complete recording is decoded in one pass rather than in chunks that can be discarded to keep a stream moving. On the retained fixture, which is generated synthetic English audio scored at three of six protected spans exact, the beginning, the negation, and the ending were all present. That is a measurement on one synthetic fixture, not an accuracy rate, and Voqela will not publish one of those until natural speech has been measured properly.

### What happens when I am not connected to the internet?

Nothing changes. Voqela is completely local: the speech model is a file on your disk, your own Apple silicon runs it, and once that one-time download is done there is nothing left for a dictation to reach out to. A plane at altitude, a train in a tunnel, a cabin, a locked-down corporate network, all the same shortcut and the same sentence. That is by construction rather than by promise, because a check fails the build on any connection-capable code anywhere in the sources outside the single model-download seam, and Settings keeps an append-only log of every network operation the app has ever started so you can read the count yourself rather than take it on faith. It has also been watched: on 3 August 2026 a transcription ran inside a process the kernel had denied the network and it succeeded, zero bytes downloaded, on named hardware. That run went through the command line against a saved recording, so what is still not on the record is the same thing done through a microphone, which is one attended test in airplane mode and a named launch gate rather than an open question.

### Do I get to keep what I said, and can my own AI read it?

Yes, and yes, with one line drawn deliberately. Every dictation is kept on this Mac in History, which is a searchable local list with Copy, Retry and an explicit Delete on the row. One transcript exports as a plain text file; the whole history exports as one JSON archive, newest first, with stable bytes so two exports can be compared. That file is yours, and handing it to whatever assistant you want reading your own words back to you is exactly what it is for. Separately, the app answers to a local command line that reports status, settings, per-application rules, diagnostics and where every file lives, with JSON on every read, so an assistant can understand your install without you narrating it. What that command line will never print is transcript text, audio, or history. Your words leave by an export you performed, and by nothing else.

### What permissions does it need, and why?

Three. Microphone, to record what you say. Accessibility, to see which field has focus and put the text into it. Input Monitoring, to notice your shortcut. Each one is explained before the system prompt appears, each has its own button, and all three can be revoked in System Settings at any time.

### What does it cost, and when can I get it?

Neither is decided. There is no price, no release date, and no distribution choice to announce, and this site will not carry one before it is real. Early access is a list, not a purchase.

### What does it need to run?

An Apple silicon Mac running macOS 14 or newer, English speech, and a one-time model download. The downloaded model artifact occupied approximately 595 MB on the Mac it was measured on.

### Is it a meeting recorder, or does it handle other languages?

No to both. Voqela is one person dictating on one Mac in English. It is not a meeting recorder, a transcription service, a mobile companion, a team workspace, or a multilingual suite, and there is no iPhone or iPad version.

## Machine endpoints

- https://voqela.com/claims.json
- https://voqela.com/agent-manifest.json
- https://voqela.com/qa-contract.json
- https://voqela.com/llms.txt
- https://voqela.com/sitemap.xml
- https://voqela.com/robots.txt