Privacy Policy

1. Overview

This Privacy Policy explains how AINSPIRED LLC ("aInspired", "we", "us", "our"), a limited liability company organized in Wisconsin, USA, collects, uses, and protects your information when you use the aInspired applications and services (together, the "Service").

Data controller: AINSPIRED LLC, Wisconsin, USA — 200 E Verona Ave PMB 5041, Verona, WI 53593. For privacy questions and data-rights requests, contact [email protected].

We designed aInspired with a privacy-first approach: keystroke data stays on your device unless you opt into encrypted cloud sync, voice recordings and text sent for correction are processed ephemerally, and we collect only what is necessary to provide the Service. Two exceptions: if you use the AI chat feature, your conversation history is stored (encrypted) so you can access it across sessions; and stickers you generate are stored and pooled into a shared catalog rather than processed ephemerally, with an opt-out — see Sections 3.2 and 3.5 for exactly how each works.

2. The Service: Our Surfaces

This one policy covers every aInspired surface. All surfaces share a single account, billing system, and backend.

• Android App — the aInspired keyboard for Android. It is currently in private testing and not yet publicly listed on Google Play; this policy applies to it today.
• iOS App — the aInspired keyboard for iOS. It is not yet released; this policy covers it prospectively. When released, it will use the same backend and the same data flows described for the Android App, and we will update this policy if its behavior differs.
• Website — ainspired.app and the backend services (account portal, admin, APIs) that power all surfaces.
• Desktop App — aInspired Desktop for Windows and Linux (macOS planned). It is not yet publicly released; this policy covers it prospectively. Detailed surface-specific data-flow disclosures for the Desktop App will be published on this page alongside its next hardened release.

Where a section below applies only to one surface, it says so; everything else applies to all surfaces.

3. Data We Collect

3.1 On-Device Data (never leaves your device)

(Android App today; iOS App when released.)

• Keystroke patterns — used by the on-device language model for next-word predictions. Processed and stored entirely on your device unless you enable optional encrypted cloud sync (see below).
• Personalized word frequencies, typo corrections, and phrase capitalizations — stored locally to improve suggestion accuracy. Also covered by the optional cloud sync described below.
• Contacts (optional permission) — if you grant the Contacts permission, the keyboard reads contact display names in memory only to capitalize the names of people you know correctly as you type. Contact data is never stored beyond the keyboard session's in-memory cache and is never sent to any server or third party.
• Notification content (optional permission) — if you grant the Notification Listener permission, the keyboard can read notification content to offer a "paste recent notification" shortcut (for example, pasting a one-time code). Notification text never leaves your device.
• Preferences and settings — theme choices, layout preferences, provider selections.
• Spell-check dictionary — local dictionary used for on-device spell checking.

The keyboard automatically excludes password fields and other secure input fields from personalized learning, and an on-device PII detection gate blocks cloud requests containing sensitive data (see Section 3.2).

Optional encrypted cloud sync: if you turn on Cloud Sync in Settings (off by default, requires Google Sign-In), the App uploads an encrypted copy of the on-device personalization data above — plus, only if you separately opt into swipe-gesture research, your swipe-typing traces — so it can be restored if you reinstall the App. This sync uses a device-generated encryption key that never leaves your device and is not known to us — this path is genuinely zero-knowledge, meaning we cannot decrypt this data even if compelled to. Restoring this data onto a second device is not yet supported (each device currently holds its own key). This is separate from chat history and context-bias personalization (Section 3.2), which use a different, operator-recoverable encryption model because we must be able to read them to provide those features.

3.2 Cloud-Processed Data

Most cloud-processed data is ephemeral — we pass it to a provider to get a result and do not keep a copy. Chat messages and AI-generated stickers are the exceptions: we store chat history so your conversation is available across sessions, and we retain generated stickers as part of a shared catalog — see Section 3.5.

• Voice recordings (ephemeral for us; kept on your device in tiers) — audio you dictate is sent to your selected or tier-assigned speech-to-text provider (full current list in Section 5) for transcription, processed in real time, and not stored by us after the result is returned. On your device, some recordings are kept so you can recover or re-transcribe them, in three tiers:
  • Recordings you save — kept until you delete them.
  • Recordings whose transcription failed — deleted automatically after 7 days, and only the 20 most recent (up to 200 MB total) are kept; older ones are removed first.
  • Recordings interrupted mid-dictation (you closed the keyboard while still speaking) and voice memos you record on purpose — deleted automatically after 30 days. Interrupted dictation is saved for the same reason a failed transcription is: so the words you already spoke are not lost.

  Voice memos are recordings you start yourself, in the aInspired app. There is a record control on the Recordings screen (Android: Settings → Voice & Dictation → Recordings; the iOS app, when released: the Recordings screen). You tap to start and tap to stop, and while it is recording your device shows its own microphone indicator (on Android there is also a notification, shown for the whole time you are recording; dismissing that notification does not stop the recording — only the Stop button does, and the microphone indicator stays up either way). The keyboard never records a voice memo — it has no way to, on either platform — and nothing is ever captured before you tap: there is no pre-roll, no always-on listening, and no recording while you are simply typing. Once you have started a memo it keeps recording if you switch apps or the screen turns off, until you stop it, and the microphone indicator stays up the whole time. A memo is saved to the 30-day tier above; you can move it to "saved" yourself if you want to keep it.

  Where a memo is transcribed differs by platform, so we say both. On Android, transcribing a voice memo happens on your device — memos are not offered to the cloud transcription providers, so a memo's audio does not leave your device by that route. On iOS (when released) there is no on-device transcription in the app, so tapping Transcribe on a recording — including a voice memo — uploads its audio to your selected speech-to-text provider, exactly as dictation does. Nothing is uploaded unless you tap it.

  Recordings also have an overall storage limit you choose in the app (2, 4, 8 or 16 GB; 4 GB by default). If your recordings pass it, the oldest failed transcriptions are deleted first, then the oldest voice memos. Recordings you saved are never deleted automatically to make room — if only saved recordings remain, the app tells you and lets you decide. The limit, its options and that eviction order are the same on Android and, when the iOS app is released, in the iOS app — the same rule, ported, so the two cannot drift apart — and on both you can change it in the app.

  Scheduled on-device transcription (Android App; off by default). You can ask the app to work through recordings you have queued for transcription starting at a time you choose. It runs on your device using on-device models only — nothing is sent to us or to any provider, and it works with no network at all. Android limits how long the app may run this way each day, so a long recording can continue in a later session; we do not promise when it will finish. If you leave the setting off, or decline the permissions it asks for, queued recordings are still transcribed on your device — in short background sessions while the device is plugged in, rather than at a set time. The app always shows which of these is in force.

  Transcripts produced from these recordings — and summaries, if you turn on on-device summarization — are stored beside the audio (saved_recordings/) and deleted with it. These files stay on your device. They are not uploaded to us; on Android they are excluded from cloud backup and from device-to-device transfer, and on iOS from iCloud backup. Audio leaves your device only when you transcribe a recording with a cloud provider; transcription that runs in the background, and summarization, use on-device models only and send nothing. They are protected by your device's app sandbox and its operating-system file encryption; we do not add our own encryption layer on top, so we do not describe them as encrypted at rest. You can delete any recording at any time from the app, and deleting your account or clearing the app's data removes all of them. Automatic deletion is a periodic sweep, not a stopwatch: on Android it runs when you open the keyboard (at most once every 6 hours) with a daily background backstop; on iOS it runs when you open the app or the recordings screen, plus a background task that iOS schedules at its own discretion and does not guarantee to run — so a recording may persist a little past its window until the next sweep, and we do not promise deletion at an exact time.

• Text for phrase correction (ephemeral) — sent to your selected or tier-assigned LLM provider (Claude, OpenAI, Groq, Gemini, Mistral, Fireworks AI, Lightning AI, Novita AI, or Xiaomi) for grammar and style corrections. Not stored by us.
• Meme prompts (ephemeral) — text descriptions sent to generate memes. Not stored after generation.
• Sticker prompts (stored) — text descriptions sent to generate AI stickers. Unlike meme prompts, these are not ephemeral — see Section 3.5 for exactly what is retained and why.
• Chat messages (stored) — sent to your selected AI provider for responses. Unlike the items above, we store your chat message content, conversation titles, and any files/artifacts created in chat on our servers so your history persists across sessions and devices. This content is encrypted at rest under a server-held encryption key unique to your account. This is an operator-recoverable encryption model, not a zero-knowledge one — we can and do decrypt this content to render your chat history, inject relevant context into your requests, and assist with support questions. Its protection is that the key is destroyed the moment you delete your account ("crypto-shredding"), permanently rendering all of your chat history unreadable — see Section 11.
• Context-bias personalization (stored, opt-in) — a separate, off-by-default feature that learns proper nouns you use often (names, places, custom terms) so speech-to-text and text correction recognize them more accurately. If you explicitly turn this on, these terms are uploaded to and stored on our servers, encrypted at rest under the same operator-recoverable model as chat messages (again, not zero-knowledge, because we must be able to read them to bias your requests). Turning the feature off, or revoking consent, deletes your stored terms from our servers.

Before sending text to cloud providers, our on-device PII detection system scans for sensitive information (credit card numbers, Social Security numbers, API keys, passwords) and blocks the request if PII is detected.

3.3 Account and Billing Data

• Google account info — email address, display name, and profile photo, collected when you sign in with Google. One account spans all surfaces.
• Device identifier — a randomly generated UUID identifying your device for authentication and quota tracking.
• Usage quotas — counts of AI correction, transcription, and chat requests, for billing enforcement.
• Subscription status — plan tier and billing cycle. In-app subscriptions are managed via Google Play Billing (through RevenueCat); web-portal subscriptions are managed by Stripe.
• API keys (BYOK) — if you bring your own provider keys, they are stored on-device only (encrypted local storage on Android, Keychain on iOS) and never transmitted to or stored on our servers. Keys do not sync across devices and are managed from within the app; removing a key is immediate and permanent.

We never see or store your payment card numbers. In-app subscriptions are purchased through Google Play Billing (via RevenueCat); web-portal subscriptions are processed by Stripe under their privacy policy. See RevenueCat's privacy policy for in-app billing.

3.4 Automatically Collected Data

• App analytics (Android App) — Firebase Analytics (Google) collects app-interaction events (screen views, feature-usage counts) and a Firebase Installation ID to help us understand feature usage and stability.
• Crash reports (Android App) — Firebase Crashlytics (Google) collects crash stack traces and device state at crash. Crashlytics retains crash data for 90 days, after which it is automatically deleted.
• Website analytics — basic page-view data via Cloudflare Analytics (no cookies, no personal data).
• Contact form (Website) — if you use the contact form at ainspired.app/contact, we collect the name, email address, subject, and message you submit. The message is emailed to us via Resend and a record is kept on our servers so we can respond and track the conversation.
• Referral links (Website) — opening a referral link (ainspired.app/r/…) validates the code and stores it in a 30-day cookie so the referral can be credited if you sign up. No other data is collected by the referral page.

3.5 AI-Generated Stickers

Unlike the meme prompts above, stickers you generate in Sticker Studio are not ephemeral — we keep them.

• Shared catalog — while sticker sharing is on (see below), every sticker you generate is added to a shared catalog and may be suggested to other people who type a similar phrase. This is by design: it is what lets us offer instant sticker suggestions without generating the same image twice.
• Your phrase is not stored as free text — the words you type to generate a sticker are sent to our image-generation and description-rewriting providers (Section 5) and used in memory to produce two separate things: a one-way hash of your exact phrase (used only to detect exact-duplicate requests; it cannot be reversed back into text), and a numerical representation of your phrase's meaning (used to find and suggest similar stickers to other people, and kept as part of the catalog). What we keep is the resulting image, a short description written by our system rather than your original wording (or, if that rewrite step is unavailable, no description at all — never your original wording as a fallback), that one-way hash, and that numerical representation.
• Consent and opt-out — generating a sticker is a deliberate action, and we show this sharing disclosure next to the Generate button before you take it; tapping Generate is the affirmative action we rely on as your consent to add the resulting image to the shared catalog (Section 12 explains how this fits our GDPR legal bases). You can withdraw that consent at any time with the "Share my stickers to the shared catalog" toggle in Settings → Privacy (on by default); turning it off stops your stickers — including ones you already generated — from being suggested to anyone else, and they keep working normally for you.
• Account deletion — deleting your account does not remove the stickers you generated. This applies to all of them, not only the shared ones: the image and the numerical representation of the phrase stay, and we remove the link between those stickers and your account and clear the description. Stickers that were in the shared catalog may still be suggested to other people; stickers you kept private stay private and are not shown to anyone.

4. How We Use Your Data

• Provide and improve keyboard predictions and autocomplete (on-device only)
• Process voice transcription requests through your selected provider
• Process text correction and chat requests through your selected AI provider
• Manage your account, subscription, and usage quotas
• Send transactional emails (billing receipts, account notifications) and respond to your inquiries
• Diagnose and fix technical issues via crash reports
• Display ads to free-tier users (Google AdMob, rewarded ads only)
• Suggest AI-generated stickers — including ones other users generated — in the keyboard suggestion strip (Section 3.5)

We do not sell your personal data. We do not use your keystroke data, voice recordings, or chat messages to train AI models.

5. Third-Party Data Processors

We share data with these services only as necessary to provide the Service:

Notification Listener: notification text read by the optional shortcut never leaves your device (Section 3.1).

BYOK users: your own API keys are stored on-device only and never transmitted to or stored on our servers. When using your own key, requests go straight to your chosen provider and your data does not pass through our servers.

6. Data Retention

• On-device data — stored until you uninstall the App or clear its data.
• Voice recordings, phrase-correction text, meme prompts — ephemeral on our side; we keep no copy after the request completes (upstream provider retention follows their own policies).
• Voice recordings kept on your device — three tiers: recordings you save, until you delete them; failed transcriptions, 7 days (and only the 20 most recent / 200 MB); interrupted dictation and voice memos you record yourself, 30 days. An overall storage limit you choose (2/4/8/16 GB, 4 GB by default) bounds all three; over it, the oldest failed transcriptions go first, then the oldest voice memos, and recordings you saved are never deleted automatically. That limit, those options and that order are the same on Android and, when the iOS app is released, in the iOS app — the Android rule was ported rather than rewritten, so one sentence describes both. Transcripts and on-device summaries are deleted with the recording. Device-local only, excluded from cloud backup, and removed by the in-app delete action, the automatic sweep, account deletion, or clearing app data. Deletion is by periodic sweep rather than at an exact time — see Section 3.2.
• AI-generated stickers (Section 3.5) — retained indefinitely as part of the shared catalog, including after you delete your account; account deletion removes only the link between a sticker and your account, not the sticker itself. Turning off sticker sharing in Settings stops your stickers from being suggested to others.
• Chat messages, conversation titles, chat artifacts — stored encrypted (operator-recoverable, Section 3.2) until you delete your account; deletion crypto-shreds the key, permanently rendering the history unreadable.
• Context-bias terms (opt-in) — until you turn the feature off, revoke consent, or delete your account; any of these deletes the stored terms.
• Encrypted cloud-sync data (opt-in, zero-knowledge) — until you disable Cloud Sync or delete your account; we cannot read it regardless of retention.
• Account data (email, name, device ID) — while your account is active; deleted within 30 days of account deletion.
• Usage quota data — retained for billing reconciliation, then aggregated and anonymized.
• Billing records — retained as required by tax and financial regulations (typically 7 years).
• Crash reports (Crashlytics) — retained by Google for 90 days, then auto-deleted.
• Contact-form submissions — retained as long as needed to handle your inquiry; deleted on request to [email protected].

7. Advertising and Tracking

Free-tier users see ads served by Google AdMob — rewarded ads only (shown only when you choose to watch one; no banners). AdMob may collect device identifiers (such as the Android Advertising ID) to serve and measure ads.

• Personalized ads require your explicit consent (collected via Google's UMP consent flow). You can opt out at any time in App settings.
• Non-personalized ads are served if you decline personalization.
• Paid subscribers (Basic, Pro, BYOK) do not see ads and no ad-related data is collected.

8. Cookies

The Website uses minimal cookies:

• Session cookie — maintains your login state; expires when you close your browser or after 30 days.
• Consent preference — remembers your cookie/ad consent choice.
• Referral code — set only if you open a referral link; stores the code for 30 days so the referral can be credited.

We do not use marketing, analytics, or third-party tracking cookies on the Website.

9. Data Security

• Encryption in transit — all network communication uses TLS 1.2+.
• Two-tier encryption at rest for synced/stored data — (1) optional cloud sync of on-device personalization data (Section 3.1) is encrypted under a device-generated key we never see and cannot recover (zero-knowledge); (2) chat history and context-bias terms (Section 3.2) are encrypted under a server-held key so we can process them on your behalf (operator-recoverable, protected by destroying the key on account deletion rather than by us being mathematically unable to read it). We do not describe tier (2) as zero-knowledge.
• On-device recordings — audio kept on your device (including voice memos you record yourself), its transcripts, on-device summaries, and the index of them are protected by the app sandbox and your operating system's file encryption (file-based encryption on Android, and the equivalent on iOS). We do not apply an additional encryption key of our own to these files, so we do not call them encrypted at rest; what protects them instead is that they never leave your device, are excluded from backup, and are deleted on the enforced schedule in Section 3.2.
• On-device PII detection — blocks cloud requests containing sensitive data.
• Secure field exclusion — the keyboard does not learn from password or secure input fields.
• HMAC token authentication — device-to-server communication uses cryptographic tokens, not plain API keys.
• Infrastructure isolation — our backend runs on Cloudflare Workers with edge-level security.

10. International Data Transfers

Your data may be processed in the United States and other countries where our providers operate. For transfers outside the EEA we rely on Standard Contractual Clauses (SCCs), adequacy decisions where applicable, and provider certifications (e.g., Cloudflare's GDPR compliance framework).

11. Your Rights

Depending on your location, you have the right to: access a copy of your personal data; correct inaccurate data; delete your personal data and account; portability (structured, machine-readable format); restrict processing; withdraw consent at any time (ad personalization, context-bias, cloud sync, swipe research, sticker sharing); and object to processing based on legitimate interest.

To delete your account and data: email [email protected] from your account email. Deletion completes within 30 days and includes crypto-shredding of your chat and context-bias encryption key, except that AI-generated stickers are un-linked from your account rather than deleted (Section 3.5).

For any request, email [email protected]. We respond within 30 days (California requests: within 45 days, as the CCPA provides).

12. GDPR (European Users)

If you are in the EEA or UK, our legal bases are: contract performance (transcription, corrections, chat, billing); consent (personalized ads, context-bias, cloud sync, swipe research, and pooling your generated stickers into the shared catalog described in Section 3.5 — for stickers, that consent is given by the deliberate act of tapping Generate with the sharing disclosure shown next to it, and withdrawn at any time via the Settings → Privacy toggle, which also un-pools stickers you already generated; every consent basis listed here is withdrawable at any time); legitimate interest (crash reporting, analytics, fraud prevention, service security, balanced against your rights); legal obligation (billing records under tax law).

• The personal data we process comes from you and from your device's use of the Service; we do not obtain personal data about you from other sources.
• We do not carry out automated decision-making that produces legal or similarly significant effects about you.
• You may lodge a complaint with your local Data Protection Authority.

13. CCPA/CPRA (California Users)

Categories collected: identifiers (email, name, device UUID); commercial information (subscription tier, purchase history); internet activity (usage quotas, app interactions, crash reports); audio data (voice recordings, processed ephemerally, not stored); chat/message content (stored, encrypted, deleted via account deletion — Section 3.2); shared sticker content (generated sticker images, linked to your account until you delete your account; turning off sharing stops other people from seeing them but keeps the link so they keep working for you — Section 3.5).

We do not sell personal information and we do not share personal information for cross-context behavioral advertising (as "share" is defined by the CPRA) — ad personalization happens only in-app with your consent, so no "Do Not Sell or Share" opt-out link is required. You have the rights to know, to delete, to correct, and to non-discrimination. Exercise them via [email protected].

14. Children's Privacy

aInspired is a general-audience service. We do not knowingly collect personal information from children under 13 (or under 16 in the EEA). If you believe a child has provided us personal data, contact [email protected] and we will promptly delete it.

15. Changes to This Policy

We may update this policy from time to time. For material changes, we will update the "Last Updated" date and notify you through the App or by email.

16. Contact Us

For privacy questions, requests, or complaints:

AINSPIRED LLC, Wisconsin, USA
200 E Verona Ave PMB 5041, Verona, WI 53593