Privacy Policy
Effective date: September 23, 2026
Toki ("the app", "we") is an AI speaking coach that helps you practice a language out loud and gives you feedback on your pronunciation, grammar, and phrasing. This policy explains what data the app collects, why, where it goes, and how you can delete it.
Data we collect and why
| Data | Why |
|---|---|
| Account info (name, email, user ID) when you sign in with Apple or Google; an anonymous ID if you use the app as a guest | To create your account, keep your practice history with it, and manage your subscription |
| Your practice speech: real-time audio of what you say into the microphone, and the transcripts and scores produced from it | To recognize what you said, assess your pronunciation, run AI conversations, and build your review queue from your mistakes |
| Learning progress (lessons completed, attempt scores, recurring mistakes, speaking minutes) | To track your progress, personalize review, and enforce fair-use limits |
| Purchase history (subscription status, not payment card details) | To unlock the features you paid for. Payments are processed by Apple; we never see your payment details |
| Practice situations you write, and the community moderation records that go with them: reports you file and creators you block | To publish and moderate learner-made situations, to keep reported content out of the community list, and to hide every situation by a creator you have blocked |
| Crash and performance diagnostics (app version, crash reports, performance data) | To fix bugs, via Firebase Crashlytics. These reports are always on and are not linked to your account |
| Usage analytics (which features you use, your account ID, your subscription tier, an app-instance identifier, and an approximate location derived from your IP address) — only if you turn it on | To see which parts of the app help, via Firebase Analytics. It is off by default and collects nothing until you turn on “Share usage analytics” in Settings; you can turn it back off at any time. It never includes voice recordings or transcripts |
Your voice recordings
Audio of your practice speech is sent through Toki’s own servers (hosted on Google Cloud) to Microsoft Azure Speech — our only speech provider — to be transcribed and scored, then discarded. No other provider receives your audio: OpenAI is sent conversation text, scenario setup, or free-production transcripts, never the recording. By default we do not store your raw audio after the request completes. We keep transcripts, scores, roleplay corrections, and progress history so your reports, review queue, and learning plan work. You may separately choose “Help improve scoring” in Settings to help us evaluate and improve scoring quality. This contribution is off by default. While it is on, successful scored practice recordings may be stored privately for at most 90 days. You can stop future contributions at any time and use “Delete contributed recordings” in Settings to remove recordings already retained. Deleting your Toki account removes them as well.
Third-party AI services
Toki’s speaking feedback and conversations are provided by two third-party AI services, Microsoft Azure AI Speech and OpenAI. Before anything is sent to either of them, the app shows a screen that names both services, lists what each receives and why, and asks for your permission. Nothing is sent to them unless you tap “Agree and continue”. If you choose “Not now”, you can still browse and listen, but speaking practice, conversations, and AI-generated practice stay off. You can give or withdraw this permission at any time with “Use AI services” in Settings; withdrawing it stops all further sending immediately.
| Service | What it receives, and how we collect it | What it is used for |
|---|---|---|
| Microsoft Azure AI Speech | The audio recorded by your microphone while you speak in a lesson or conversation (collected only while the microphone button is active), plus the tutor’s lesson lines as text | Turning your speech into text, scoring your pronunciation, and producing the tutor’s voice |
| OpenAI | Text only, never audio: the transcript of what you said or the reply you typed in a lesson or conversation; the practice situations and lesson topics you write; your level, learning goals, interests, and app language; and recent mistakes from your practice history | Replying as your conversation partner, checking your free-speaking answers, explaining your mistakes, creating practice situations, lessons, and illustrations for you, and screening that content for safety |
Toki never sends your name, email address, contacts, or account identifiers to either service. Requests go through Toki’s own servers, so the services do not receive your device’s IP address from the app. Both act as our processors: under their API and data-processing terms they process this data only to provide the service to Toki, keep it confidential and secured, and do not use it to train their models. Azure AI Speech does not store your audio after processing it. OpenAI may keep API requests for up to 30 days solely to detect abuse, and then deletes them. These commitments give your data the same or equal protection as this policy.
No ads, no selling data
Toki shows no ads, does not sell your data, and does not use your data for third-party advertising or tracking across other apps.
Service providers
We share data only with the processors that make the app work:
| Provider | Used for |
|---|---|
| Microsoft Azure Speech | All three speech jobs. It receives the audio you record, for speech recognition and pronunciation scoring, only after you agree in the app (see Third-party AI services), and it receives the tutor's lines as text to synthesise the tutor voice (text-to-speech) |
| OpenAI | The AI conversation partner, free-production judge, and practice generator, only after you agree in the app (see Third-party AI services). Receives conversation and practice text as described above, and returns replies, feedback, or generated practice. It is never sent your audio |
| Google Firebase / Google Cloud | Sign-in, app attestation (App Check), the database that holds your account, learning history and community moderation records, crash reporting (Crashlytics), and the servers your practice audio and conversation text pass through on their way to Azure Speech and OpenAI — and, only while you have the Settings toggle on, usage analytics (Analytics) |
| RevenueCat | Subscription status, and the app user ID it is attached to |
| Apple | Payments and subscription purchases (StoreKit); we never see your payment details |
Each provider processes data on our behalf under its own security terms.
Data retention and deletion
Your learning history is kept while your account is active so your progress and review queue keep working. You can delete your account at any time in Profile → Settings → Delete account: this removes your account, learning history, mistake bank, contributed recordings, and the community moderation records held under your account (the reports you filed and the creators you blocked) from our systems. Practice situations you wrote and published to the community are unlinked from you and hidden from the app rather than erased, so that other learners’ saved copies and our moderation history stay intact; situations you kept private, or deleted yourself before deleting your account, are removed. If your account is linked to Sign in with Apple, we also ask Apple to revoke Toki’s Sign in with Apple authorization; your Toki account is deleted whether or not Apple confirms that, and you can always remove Toki yourself under Settings → your name → Sign in with Apple on your device. Guest accounts that are never linked may be removed after long inactivity.
Children
Toki is not directed at children under 13, and we do not knowingly collect data from them. If you believe a child has created an account, contact us and we will delete it.
Changes
If this policy changes materially, we will update this page and note it in the app. The effective date above always reflects the current version.
Contact
Questions or requests: nvkhoe89@gmail.com. See also our Terms of Use.