KeySpeech

Formerly Speakli.

A keyboard you talk to, in two styles: one big Record button, or a full keyboard with letters and dictation in a bar above them. Either way the text lands in whatever app you were typing in — and meetings you record become transcripts by speaker and summaries.

Privacy Policy

Last updated: 13 October 2026

KeySpeech is made by Pavel Demyanov. This policy explains what the app does with your data. The short version: KeySpeech has no server, no account and no analytics, and it collects nothing about you.

What stays on your device

• Your speech. When you use one of the on-device engines (KeySpeech Engine or Whisper), audio is processed on your iPhone and never leaves it. Apple's built-in engine recognises speech on the iPhone when the language is available offline; otherwise Apple may process the audio on its servers, as with the system's own dictation.
• Your text. Dictation history (the raw and processed text, and the recordings of the latest dictations so you can listen to them or recognise them again — 20 by default, 50, 100 or none in Settings › Data › Voice recordings), notes and their audio recordings, transcripts and summaries are stored only on your device, in the app's own container. They are not uploaded anywhere and are deleted when you delete them or remove the app; you can also choose how long History is kept (a week, a month, a year or forever). Like other app data, they are part of your device backup.
• Which app you dictated into. History notes the app each keyboard dictation went into, such as Mail or Telegram, so the Dashboard can show where you dictate. KeySpeech learns it on your iPhone from the system's keyboard state. It is stored with your History on the device, is never sent anywhere, and is included when you export History.
• Imported audio. When you import audio or video files into Notes from the Files app, KeySpeech keeps its own copy of the sound in the app's container on your iPhone, next to the note, and deletes it with the note. The file is transcribed with the engine you chose in Settings, so nothing is uploaded unless that engine is OpenAI. Your original file stays where it was.
• Speakers and voices. KeySpeech can tell apart the voices in a note. Unless you choose OpenAI for a note (below), this runs entirely on your iPhone, with a model you download once from Hugging Face on request (only the model files are downloaded; nothing is sent). The people you name and, if you keep "Remember voices" on, a numeric voiceprint of each named person are stored only on your device, in a folder excluded from backups, so that KeySpeech can suggest who is speaking in later notes. It never names anyone without asking. You can forget one person's voice or all voices at any time in Settings › Speakers and voices; deleting a person removes their voiceprint.
• The on-device AI model. If you download it (Settings › AI processing › Connections › On iPhone, about 3 GB, from Hugging Face on request — only the model files are downloaded; nothing is sent) and set a template to run On iPhone, that template's work — summaries, the processing of new notes, their titles — is done on your iPhone and nothing leaves it. The model works only while the app is open and is not used for dictation from the keyboard.
• Your settings and API keys. Settings live on the device. API keys are added in Settings › AI processing › Connections and stored in the iOS Keychain on that device only; they are never shared with the keyboard extension, never synced, and never sent anywhere except to the provider they belong to.

What we collect

Nothing. KeySpeech has no analytics, no crash reporting service, no advertising identifiers and no user accounts. The developer receives no data from the app.

Optional services you choose

These features send data to a third party, and only when you use them with your own API key:

• OpenAI speech recognition — with the OpenAI engine selected, the audio of a dictation, or of a note you record, import or recognise again with it, is sent to OpenAI for transcription.
• Speaker separation through OpenAI — when you choose OpenAI to separate the speakers of a note (Redo note), the note's audio is uploaded to OpenAI under your key. Before a note's audio goes to OpenAI for the first time, KeySpeech asks you; until then it stays on your iPhone.
• Templates — every dictation and summary template has its own provider and model. When a template processes text — a dictation, a new note, a note you process again, a summary or an AI title — the text is sent only to that template's provider (OpenAI, Anthropic, Google Gemini, or an OpenAI-compatible server you configure, including one running on your own computer) so it can be cleaned up, rewritten, translated or summarised as the template asks. A template that runs On iPhone sends nothing, and with AI off no text leaves the device.
• Summaries of notes — when you ask for a summary (meeting notes, a medical report or any other template), the note's whole transcript, with the names and roles you gave the speakers, is sent to the provider of that summary template. Before the first summary — or the first reprocessed note — goes to a cloud provider or to a server on the internet, KeySpeech asks for your confirmation; a server on your own network is not asked about. Conversations you summarise may contain health or other sensitive information: only create summaries of them with a provider you trust with that information.
• Note titles — with "Titles by AI" on (Settings › Notes), a new note is titled from its text (for a long note, its beginning, middle and end). The text goes only where the note's text goes anyway: to the provider of the "New notes" template; or, when new notes are not processed, to the default template's provider, and only if you have already allowed summaries to go there. A template that runs On iPhone or on a server on your own network needs no such permission. Otherwise the title is the note's first words and nothing is sent.

These requests go directly from your iPhone to that provider under your own key, on their terms and privacy policy. KeySpeech is not an intermediary and keeps no copy. None of these features runs until you enable or start it, and the app states on screen where the data goes.

Your Dictionary. Words you add to the Dictionary tab are stored on your device like the rest of your data. When OpenAI speech recognition or a template with a cloud provider is used, those words travel with the request — as a vocabulary hint sent to OpenAI together with each recording (the speaker-separation model gets none), and as a glossary added to the template's prompt — so that the provider spells them the way you asked. Otherwise the dictionary never leaves the device.

Exports. When you export a note, a summary or a transcript, the file is created on your iPhone and goes only where you send it with the iOS share sheet.

The keyboard and Full Access

KeySpeech has two keyboards — the voice keyboard with one big Record button and a full keyboard with letters — in one keyboard extension, which requires "Allow Full Access" in iOS Settings. iOS does not permit a keyboard extension to use the microphone, so the keyboard asks the KeySpeech app on the same device to record: through a container shared with the app on your iPhone, it passes on your Record and Stop taps and the template, provider and model you pick with its chips, and reads the app's status and the recognised text. Without Full Access the extension cannot reach the app at all. Full Access is used for that single purpose. The keyboard contains no network code and never sees your API keys.

The full keyboard types the characters you tap into the text field through iOS, and works out which letter you are likely to type next on the device itself, with iOS's own spell checker. It keeps only a count of the emoji you use, for Frequently Used, in its own settings on your iPhone. KeySpeech does not log, store or transmit what you type.

Microphone

KeySpeech keeps the microphone on standby in the background for the period you choose — 5, 15 or 30 minutes, 1 or 2 hours — so the keyboard can start recording instantly. You can pause standby on the Dashboard at any time, which releases the microphone. Nothing is recorded until you tap Record. A note keeps recording with the screen locked, for up to four hours, until you stop it; a Live Activity on the Lock Screen shows that it is recording and lets you stop it. While the microphone is on standby, iOS shows an orange dot in the status bar.

Children

KeySpeech is not directed at children and collects no personal information from anyone.

Changes

If this policy changes, the updated version will be posted on this page with a new date.

Contact

Questions about the app or this policy: greypax@gmail.com