Privacy Policy

1. Who we are

TalkQuill ("TalkQuill," "we," "us") makes a system-wide dictation app for Windows. This policy explains what data we process when you use the app and our website, why, and the controls you have. If you have questions, contact us at privacy@talkquill.app.

2. What we collect

Voice audio (transient)

When you trigger dictation, the audio you record is captured locally and sent to a speech-to-text provider to be converted to text. We do not retain this audio. It is processed for transcription and then discarded. It is never stored on our servers and never used to train any model.

If a dictation cannot reach the service (for example, you are offline or the connection drops mid-sentence), the audio is kept only on your device, in a local recovery folder, until it can be transcribed; it is deleted the moment transcription succeeds, and you can discard it manually at any time from History. It never goes anywhere except the same speech-to-text flow described above. Turning on Privacy Mode keeps this recovery buffer in memory only, never on disk.

Transcribed and formatted text (transient)

The raw transcript is sent to a formatting provider to remove filler, fix punctuation, and apply per-app styling. The cleaned text is returned to your machine and pasted into your active app. Unless you enable the AI-training opt-in described in section 5, we do not retain the content of your dictations on our servers.

Context Awareness hints (transient)

When Context Awareness is on and you have seen the in-app disclosure, TalkQuill may read bounded nearby text from the active field using Windows accessibility APIs. These hints can include nearby text before or after the cursor, selected text, the window title, and the focused field name. TalkQuill uses them only while you dictate to improve names, code identifiers, spelling, casing, and formatting. Password fields, blocked apps, and Privacy Mode skip expanded context. Context Awareness does not capture screenshots, record your screen, or run OCR.

Read-aloud selections (transient)

When you use Read aloud, TalkQuill captures the text you selected using Windows accessibility APIs or a temporary clipboard copy. Before upload, the app removes email transport headers, email addresses, links, file paths, likely credentials, and private-key blocks. Password fields and common password-manager apps are blocked. The cleaned selection is sent to TalkQuill and a text-to-speech processor only to create audio, then discarded by TalkQuill. It is not added to your History or training corpus. The speech processor does not use API content to train its models by default, but may retain request data temporarily for abuse monitoring under its service terms.

Account & billing data

If you create a paid account, we process your email address and billing details (handled by our payment processor). We store your subscription status to provide the service.

Usage metering (accounts)

When you use TalkQuill with an account, we record daily usage totals (words transcribed, seconds of audio processed, and request counts) and per-request technical records (timing, status, and processing cost). We use these to enforce plan limits, such as the free tier's daily word cap, and to keep the service reliable. These records never include your audio or the text of your dictations.

Local data and optional account sync

Your dictation history, custom dictionary, snippets, macros, and Voice Profile analytics are stored locally. Turning on Privacy Mode keeps dictation history off-disk entirely and pauses account sync.

If you explicitly enable account sync, up to 200 recent History entries (including raw and cleaned transcript text) and a per-device Voice Profile contribution are sent over TLS, encrypted by TalkQuill before database storage, and associated with your signed-in account. Custom dictionary entries, snippets, and macros are not synced. Sync is separate from AI-training consent. Turning sync off stops future syncing but does not delete either local data or the remote copy; use Delete remote data to remove the remote copy.

Diagnostics & telemetry (opt-in, off by default)

Content-free usage diagnostics (crash reports, latency metrics, feature counts, app version, and word-count aggregates) are off by default. They are only collected if you explicitly opt in, and never include the content of your dictations, voice audio, vocabulary, snippets, or the app you were typing into.

3. Service providers

TalkQuill is a managed service: we run the pipeline, we hold the keys, and your data moves only where it must to come back to you as text. To deliver that, we use a small number of vetted, specialist processors under contract:

• Speech-to-text processing. Receives your voice audio for the sole purpose of converting it to text. It may also receive your custom vocabulary and derived Context Awareness terms to bias recognition. Audio is discarded after transcription and never used to train models.
• AI text formatting. Receives the transcript for the sole purpose of cleaning and styling it. When Context Awareness is active, it may receive bounded nearby text hints so the returned text matches the active field. Content is not retained and never used for training.
• Text-to-speech processing. Receives only the cleaned read-aloud selection to return spoken audio. API content is not used to train the processor's models by default; limited abuse-monitoring retention may apply under the processor's service terms.
• Payments. Our payment processor handles your billing details; we never see or store your full card number.
• Hosting & delivery. Runs our servers and website. All traffic is encrypted in transit.

Processors are bound by confidentiality, purpose limits, and applicable data-processing terms. TalkQuill does not permit them to use your dictation or selected text for advertising or model training. Provider-specific security and abuse-monitoring retention can apply where disclosed above. Enterprise customers can request the current processor list under NDA as part of a Data Processing Agreement.

4. How we use data

• To transcribe your speech and return clean, formatted text.
• To provide, maintain, and secure your account and subscription.
• To create natural speech from a cleaned selection you choose to read aloud.
• To improve reliability and performance, only via opt-in, content-free usage diagnostics.
• To enforce plan limits and prevent abuse, using content-free usage totals.

We do not sell your personal data. We do not use your dictation content for advertising. We do not train models on your content unless you affirmatively opt in (see section 5).

5. AI-training opt-in

TalkQuill ships with the AI-improvement opt-in disabled. We do not train on your content unless you affirmatively enable this setting, which you can turn back off at any time.

If you enable it, each successful dictation contributes its raw transcript and final formatted text (never audio) to a corpus used to train TalkQuill's own formatting engine. Contributions are tagged with a random contribution ID generated on your device and, on accounts, associated with your account ID. Contribution pauses automatically while Privacy Mode is on and stops as soon as you turn the setting off. You can have everything you previously contributed deleted at any time by emailing privacy@talkquill.app.

6. Your controls & rights

• Export. Export local History and request an in-app export of diagnostics, training contributions, and synced account data.
• Delete. Clear local History, propagate a synced-history deletion, or use Delete remote data to remove diagnostics, training contributions, synced History, and synced Voice Profile data.
• Privacy Mode. Keep history off-disk entirely, on any plan including the free tier.
• Context Awareness. Turn nearby-text context off in Settings at any time. Privacy Mode pauses expanded context automatically.
• Training contributions. Turn the AI-training contribution off at any time; on request we delete everything you previously contributed.
• Access & correction. Subject to applicable law (including GDPR and CCPA), you may request access to, correction of, or deletion of personal data we hold. Email privacy@talkquill.app.

7. Security

Data in transit is encrypted with TLS, and access to production systems is limited to what operating the service requires. Synced History and Voice Profile data also receive application-layer encryption before database storage. This is encryption at rest, not end-to-end encryption: TalkQuill's service holds the encryption key so it can merge and return your account data. Local data is protected by your operating system's user account controls.

8. Data retention

Voice audio is not retained beyond transcription. Read-aloud selections and generated audio are not persisted by TalkQuill; the speech processor's limited abuse-monitoring retention may apply as described above. Synced History and Voice Profile data remain until you use Delete remote data or delete the account. Individual sync deletion markers, without transcript content, can remain for up to 90 days so an offline device cannot restore deleted history. Training contributions remain until deleted. Content-free request diagnostics are retained for up to 90 days, processed billing-event identifiers for up to 180 days, and anonymous operational aggregates for up to 400 days. Usage metering, account, and billing records are kept while the account is active and as required for legal and accounting obligations.

9. Children

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

10. Changes

We may update this policy. Material changes will be announced in-app and reflected in the "last updated" date above.

11. Contact

Privacy questions, data requests, or security reports: privacy@talkquill.app.