Privacy Policy

SpeakMind is a macOS dictation app: you hold a hotkey, speak, and it types clean text wherever your cursor is. This policy explains exactly what data SpeakMind handles, where it goes, and the choices you have.

We designed SpeakMind to keep your words yours. We do not sell your data and we do not show ads. The website uses privacy-conscious Vercel Web Analytics to count page views and interactions with product examples, help, language selection, access guidance and checkout links; it does not create advertising profiles or inspect anything you dictate in the app. URL query parameters and fragments are removed before website analytics events are sent. The website respects browser Do Not Track and Global Privacy Control signals.

1. How your words flow

Every dictation follows the same short path. Two cloud AI providers can be involved — one to turn your speech into text, another to clean that text up:

• 1 · Capture. Audio is recorded only during an active recording session. Hold and release the dictation hotkey, or use a start/stop control, to manage recording.
• 2 · Transcription (speech → text). By default the audio is sent through SpeakMind's secure proxy to OpenAI (gpt‑4o‑transcribe), which turns it into raw text. Turn on On-device Mode to use Apple's on‑device recognition instead — then audio never leaves your Mac.
• 3 · Cleanup (text → clean text). The raw transcript is sent to a language model to remove fillers, fix punctuation and format it. By default this runs on Cerebras (our primary cleanup provider), with OpenAI as a fallback. A China‑domestic route to Volcengine (Doubao) also exists for lower latency and is used only where enabled.
• 4 · Insert. The finished text is typed at your cursor. Provider retention depends on the service and account configuration; each processor handles data under its own terms.

Bring your own key (BYOK): point transcription and cleanup at your own provider account (OpenAI, Groq, Cerebras, OpenRouter, Gemini, DeepSeek, or a custom endpoint). Your audio and text then go directly to that provider, not through our proxy. Ask mode: when you ask a question by voice, your question is sent to OpenAI's Responses API and may trigger a web search to answer it.

2. What SpeakMind processes

Your voice (audio)

Audio is captured while a recording session is active, using hold-to-talk or a start/stop toggle, and is used only to produce your transcript (the pipeline above shows exactly where it is sent). Cloud-provider retention depends on the service and account configuration.

On your Mac, a recording is written during each capture. Depending on your History settings, a copy may be kept locally so you can re‑transcribe it later; you can turn this off and delete recordings any time in Settings.

Your transcribed text (cleanup)

After transcription, the raw text is sent to a cleanup language model to remove filler words, fix punctuation and format it. By default this is Cerebras, with OpenAI as a fallback (and Volcengine / Doubao on the China route, where enabled). Cleanup is faithful — it tidies what you said, it does not invent content. You can lower the cleanup strength or use lighter modes in Settings to keep more work on‑device.

Your account / billing

• If you enter an email to check your subscription, that email is sent to our license service (backed by Stripe for payments) to look up your plan. Payment card details are handled entirely by Stripe — SpeakMind never sees or stores your card.
• A randomly generated device identifier is used to manage your free trial and usage quota.

Stored on your Mac (local only)

The following stay on your device (in macOS preferences / Keychain) and are not uploaded by SpeakMind:

• Dictation history and the text you've dictated (you can disable history, and set auto‑deletion, in Settings).
• Your Memory Dictionary, snippets, transforms, and app settings.
• Any API key you provide (stored in the macOS Keychain, device‑only).

Website early-access requests

When the online request form is available, it stores the email address you submit, your selected website language, the privacy-notice version and the request timestamp in private Supabase storage. These details are used to manage your early-access request and contact you about access. The website does not create an App account, start a trial, send a confirmation email, or subscribe you to marketing messages when you submit this form.

To limit automated submissions, the server derives a keyed hash from the connection IP address; the raw IP is not stored in the request database. Rate-limit records are cleaned as new requests are processed. Application code does not log submitted emails, raw IP addresses or database error details. The hosting provider may maintain its own request logs. If you want to withdraw your request or have its record deleted, contact hello@speakmind.app from the address you submitted.

Submitting the same email again preserves the original request. A saved-request message does not confirm an invitation or download. If online storage is unavailable, the page offers email contact instead; opening a mail draft alone does not submit a request.

Diagnostics

SpeakMind can capture local crash logs on your Mac. If you choose to export a diagnostics report, it is written to your Downloads folder for you to share manually — emails and account tokens are stripped from that report before it is written. SpeakMind does not automatically transmit crash logs.

3. Companies that process your data

Each processor handles data under its own privacy terms. We send them only what is needed to provide the feature you invoked. If you use Bring‑your‑own‑key, your audio and text go directly to the provider you choose instead of the defaults above.

4. Your choices

• On-device Mode (on‑device): keep transcription on your Mac with Apple's on‑device recognition, so audio never reaches a cloud provider.
• Bring your own key (BYOK): route transcription and cleanup through your own provider account (OpenAI, Groq, Cerebras, OpenRouter, Gemini, DeepSeek, or a custom endpoint) instead of our proxy.
• Reduce cloud use: lower the cleanup strength or use lighter modes to keep more processing on‑device.
• History off / auto‑delete: disable saving, or auto‑delete older entries and recordings, in Settings.
• Permissions: SpeakMind records only during a session you start and uses Microphone, Speech Recognition, and Accessibility permissions, which you grant in macOS and can revoke any time in System Settings.

5. Security & data retention

• Audio: cloud-provider retention depends on the service and account configuration. On your Mac, recordings are kept only as long as your History settings allow — disable history or set auto‑delete to remove them.
• History / dictionary / settings: kept locally until you delete them (or per your auto‑delete setting).
• Account/billing: retained by Stripe per its policies for as long as you have an account.
• Keys & secrets: any API key you provide is stored in the macOS Keychain on your device only; traffic to our services and processors is sent over encrypted connections.

6. Children

SpeakMind is not directed to children under 13 (or the minimum age in your region) and we do not knowingly collect their data.

7. Changes

We may update this policy; the "Last updated" date above will change and, for material changes, we will surface a notice in the app.

8. Contact

Questions or data requests: hello@speakmind.app.

SpeakMind is operated by Kanverse APP, based in Sydney, Australia.