Who this policy covers
EnterSpeak is provided by Dayella Limited. This policy describes the current EnterSpeak apps and this static website, including the private beta. No EnterSpeak account is required for local dictation. This policy will be updated here if the service changes.
Local processing and recordings
When you select on-device processing, compatible speech, speaker-identification, handwriting and writing models process content on your device. This app’s Apple Speech route requires on-device recognition and does not fall back to server recognition when it is unavailable. Translation in updated app builds follows the text model selected in Processing mode: on-device translation uses the downloaded model and does not require Apple Intelligence; selecting your own API sends the transcript or entered text to the disclosed endpoint only after separate translation consent. Older Apple app builds can use Apple TranslationSession with system-managed local language resources. Model and language-resource downloads are separate network operations. These statements describe EnterSpeak’s configured routes, not every Apple dictation or translation feature.
On Apple platforms, dictation audio is temporary and normally removed after processing; interruptions can leave files until cleanup. Saved meeting and translation recordings remain until you delete them. Imported source files remain in their original location. History, original and corrected text, preferences and dictionary entries are stored locally while you keep them. Saved corrections do not retrain the local model.
Optional managed cloud AI and bring-your-own API routes transmit the data described below. Local inference does not make model downloads, optional iCloud sync, purchase verification or support chat offline. You can use available on-device features without a cloud AI account or remote-AI permission; device, language and model requirements still apply.
Choose what may leave your device
In the Apple apps, remote-AI permission is an explicit choice after a disclosure of the feature, data and recipient. Managed cloud audio and managed cloud text have separate permissions. Your own API has separate speech, writing and translation permissions, tied to the saved connection, model and disclosed destinations. Optional speech recognition hints require an additional choice. Saving an API key, signing in, purchasing access, allowing the microphone or enabling keyboard Full Access does not grant these permissions.
Permissions record the disclosure version and recipient or connection. A changed disclosure or recipient, or a changed API connection or model, requires review again. Previously saved settings do not count as consent to a new disclosure. Declining leaves available local features accessible. Returning from a permission review does not start microphone capture automatically; start or resume the operation yourself.
Withdraw managed cloud permission in AI Credits, or API permission in the relevant API settings. Withdrawal blocks subsequent sends and retries on this device. For managed cloud work already queued on our server, the app must deliver the withdrawal while connected and signed in; it retains an undelivered request for retry. Server work may continue before delivery. A withdrawal received before provider dispatch prevents that queued work from being sent. Neither withdrawal, Stop/Cancel nor deleting a local result can recall a request already sent to a provider or guarantee deletion of its copies. Granting permission again does not revive an old operation.
EnterSpeak managed cloud AI: data and recipients
If you select managed cloud speech, EnterSpeak sends the selected microphone or imported audio in short WAV clips, with language and processing settings and a request identifier, to Dayella’s EnterSpeak service hosted on Cloudflare Workers and D1. The service sends the audio to BytePlus Seed ASR for transcription. Depending on the speech mode, BytePlus receives the audio bytes or a temporary signed URL from which it can retrieve the clip. This audio route can be used by dictation, keyboard voice, imports, meeting transcription and live-caption recognition. It does not send your saved dictionary as recognition hints.
If you select managed cloud writing, EnterSpeak sends the text or transcript, your reply instruction where applicable, selected language/style/domain, and relevant preferred wording, correction references or custom writing instructions through the same EnterSpeak service to BytePlus ModelArk for text processing (DeepSeek V4.1 Flash). Content may identify you or other people. Cloudflare hosts request data, results and service records; it is not the model provider for these routes.
Audio recognition and the next text operation are separate choices. Current Apple translation uses Apple’s framework or the separately selected API translation route, not the managed writing route. Meeting notes use local processing unless you choose your own API for the summary; selecting managed speech can still send the meeting audio for transcription.
Your own API: destinations and data
When you choose your own API, content goes to the endpoint you configure, including any gateway and its downstream model provider. Speech services can also use a separate upload service: AssemblyAI uploads audio before submitting the recognition job, and Seed ASR Standard uses the separately configured audio-upload destination. Review every displayed destination and the actual service/account terms before agreeing. API credentials identify your provider account, and the destination receives network/request metadata. The app does not add your Apple sign-in identity to these inference payloads.
Speech: selected WAV audio clips, model/language settings and required credentials. For formats that support it, separately permitted recognition hints can include your spoken-language profile and up to 12 preferred terms from your saved vocabulary. Withdrawing hints stops subsequent use of that optional scope; audio permission is separate. Writing and meeting notes: selected text, transcripts or intermediate summaries, instructions, language/style/domain settings, and relevant preferred wording, correction references or custom instructions. A meeting-summary text request does not itself upload the recording.
Translation: source text and source/target languages; original text and the proposed translation may be sent again for meaning verification. Writing and translation validation can make additional requests. Voice translation may first require separate speech processing and audio permission. Checking API model options: sends credentials, model identifiers and network metadata after a separate acknowledgement, without a recording or transcript; it does not grant content-processing permission.
On Apple devices API keys are stored in device-only Keychain. Android API keys use Android Keystore encryption and are excluded from backups. Removing a connection or returning to local mode stops its use in the app; it does not erase data already held by the provider. Retention, training, human review, processing locations, deletion and fees depend on the actual provider, gateway, upload service and your account agreement. We do not promise that every configured provider has the same policies or protections.
Cloud storage, provider terms and deletion limits
Our managed service temporarily stores audio or text inputs and processing options. Input records are automatically removed from the live database when a job finishes or fails. Remaining inputs expire four hours after upload. An independent cleanup job is scheduled every minute, including when cloud recognition is disabled, and clears returned text 24 hours after its last update. Successful cleanup records an operational timestamp; detected cleanup failures trigger operational alerts. Outages can delay cleanup. These are service cleanup thresholds, not guaranteed physical-deletion deadlines or provider-retention limits. Scheduling, interruption, recovery copies and infrastructure logs can have different lifetimes. Clearing a result in the app or deleting a live database row does not establish that every backup or provider copy has been erased.
Account, request/job and transaction references, duration or token counts, status/error codes, timestamps and credit/usage records support service delivery, purchase verification, security, abuse prevention and accounting. These records have different retention needs from audio/text inputs and may remain after content or account deletion where needed for those purposes or legal obligations. We do not state a single fixed deletion period for all such records. Cloudflare and inference services also process network and operational metadata.
BytePlus ModelArk data processing information and its standard online terms apply to cloud text processing. We do not promise zero retention or a fixed deletion period. Data may be retained for service delivery, security and legal requirements. BytePlus’s published Seed Speech and Vision AI terms cover speech-to-text and incorporate its General AI Terms. Those terms describe retention for service delivery, maintenance, fraud/abuse prevention and security, including abuse review. They limit foundation-model training, with exceptions for voluntary feedback, separate consent or written agreement. We use the standard online terms; we do not promise zero retention, no review, or a fixed per-request deletion period. The precise version applicable to an existing account depends on acceptance and transition terms. Cloudflare’s data-processing terms describe its infrastructure role; they do not by themselves prove a particular backup/log deletion time. Processing may occur outside your country. Contact us below for questions about a recipient or a deletion request.
Android recordings, keyboard, translation and backup
The Android app processes speech, speaker grouping and supported writing locally by default. Saved dictation, meeting and translation recordings, transcripts, preferences, dictionary entries and cumulative usage statistics remain on your device until you remove them. This differs from temporary dictation audio on Apple platforms.
If you separately enable and consent to ASR API, selected microphone or imported audio is sent in short clips to your configured transcription provider. Optional writing, translation and meeting-note APIs send the relevant text, selected settings and, for notes, intermediate summaries. The provider follows its own retention and privacy rules. Meeting-note processing alone does not upload the recording. API keys are encrypted with Android Keystore, excluded from backups and removable in AI Setting.
The Android app uses Google ML Kit for Chinese digital-ink handwriting and on-device language identification. Translation uses the local text model selected in Processing mode; this release does not use ML Kit translation. The SDK may contact Google for model downloads and configuration and collect technical diagnostics and usage analytics, including device/app information, installation identifiers, performance and API events, configuration, input/output sizes and identified languages. These metrics are separate from recordings and transcript text; the app does not provide an effective switch to turn off all SDK metrics. See Google’s ML Kit data disclosure and Google’s privacy policy.
The Android keyboard uses microphone permission for a recording that you start and a visible foreground service while you return to another app. It reads bounded surrounding editor text locally to check the insertion destination. That surrounding context is not saved in history or sent to API providers. Stop/Cancel ends microphone capture. Apple keyboard Full Access is an Apple-specific permission, not an Android requirement.
For app-controlled transmissions, EnterSpeak requires HTTPS: managed cloud AI requests, Google Play purchase verification, custom API connections and model downloads are rejected unless they use HTTPS, and the Android app disables cleartext HTTP traffic. This covers transport from the app to those endpoints only. It does not cover user-directed exports or what destinations outside EnterSpeak’s control do after receiving data; those destinations apply their own security practices, and exported ZIP files are not encrypted by EnterSpeak.
Android automatic operating-system cloud backup and device transfer are disabled. Android does not currently use the Apple iCloud sync described below. You can explicitly export an Android backup, optionally including audio, to a location you choose. The exported ZIP is not encrypted. It includes saved text, notes, captions, preferences and usage statistics; API keys, model weights, paid-access receipts and pending keyboard text are excluded. Other apps and storage providers handle exported copies under their own rules.
You can delete saved recordings/text and remove your API connection in the app, or revoke microphone/keyboard access in Android settings. Removing local data or uninstalling does not delete exported copies or data retained by API/SDK providers. Google Play purchase and restoration use the verification process described below. Manage or cancel subscriptions in Google Play; removing app data does not stop renewal.
Optional Google sign-in creates a Dayella cloud account. We verify the Google identity token and store its issuer, user identifier, internal account reference, sessions and wallet records. We do not store your Google password. Local dictation and Google Play Plus/Pro access do not require this cloud account. Managed cloud speech, managed writing and credit top-ups are not promised for every installed version or device; availability varies by installed app version and the service status shown in the app. Where offered, managed speech and writing each require separate explicit consent, sign-in, a network connection and available AI Credits. Selecting managed writing with on-device speech sends the relevant text, instructions and settings, not the recording through the writing request, to the recipients described under “EnterSpeak managed cloud AI”. Google Play Plus/Pro access is separate from AI Credits, and buying credits does not grant remote-AI permission.
Delete an Android cloud account: in EnterSpeak, open the cloud AI account controls and choose “Delete cloud account”. Confirm deletion and forfeiture of the remaining credit balance. This revokes sessions, removes stored request content and payloads, and removes the Google identifier from the account. Internal account references and linked transaction or usage records remain for accounting and prevention of duplicate purchase claims; removing the Google identifier is not irreversible anonymisation. Local recordings and exported copies must be deleted separately. Deletion does not cancel Google Play subscriptions.
Without the app, open the EnterSpeak support conversation and request “Android cloud account deletion”. Provide only the account reference or contact information needed to verify ownership. Do not send passwords, identity tokens or purchase tokens. We verify ownership before deletion and explain records that must be retained.
Microphone, keyboard and clipboard
Microphone permission is used for dictation in the main app, including an active background microphone session used by the keyboard. Keyboard Full Access allows communication with the main app using shared on-device storage. Dictation commands and results pass between the app and keyboard there.
Copy actions place the selected text on the device clipboard. Keyboard insertion sends the result to the text field in the app you are using; that destination app handles it under its own privacy practices. Avoid dictating information you do not want that app to receive.
Optional private iCloud sync — Apple platforms
If enabled with eligible access, supported history text, record identifiers, timestamps, deletion markers, usage statistics and supported mode/language preferences sync through your private Apple iCloud storage. Audio recordings and AI model weights are not uploaded by this sync. Dayella has no general administrative access to your private CloudKit content.
Sync requires your iCloud account and a connection. Turning sync off does not erase existing cloud copies. Apple operates iCloud under its privacy policy.
Model downloads and service providers
Model files download from their providers and delivery networks, including Hugging Face. Those services receive network information needed to serve the request, such as your IP address and requested file. Downloading a model does not upload your recordings or transcripts to those providers.
Apple provides app distribution, optional iCloud storage and purchase services. Cloudflare hosts and protects this website and processes network request information. These services have their own privacy policies: Apple, Hugging Face and Cloudflare. Their processing locations may differ from your country.
Cloud accounts, purchases and service records
Apple processes Apple-platform payments. EnterSpeak verifies transactions and access status; payment-card details and Apple Account email are not supplied by that verification. Optional Sign in with Apple requests no name/email scopes. Signing in to the managed cloud service creates an EnterSpeak account: we verify the identity token and keep its issuer/user identifier, an internal account reference, sessions and wallet records. We do not receive your Apple password. Local dictation does not require this account.
Managed cloud requests use an authenticated session and request identifier. The service records purchase references, product identifiers, credit movements, processing duration or token counts, status and timestamps for access, accounting and abuse prevention. Network infrastructure can receive your IP address and app/request metadata. The service also records the permission disclosure version, recipient, grant identifier and dispatch or withdrawal times to enforce permissions and audit requests. These records are separate from local history and from the temporary content described above.
Local usage counts remain on the device; supported statistics can sync when you enable private iCloud sync. The weekly allowance ledger is separate from history, so deleting a transcript does not reset it.
Google Play handles Android payments. When you purchase or restore a Google Play plan, EnterSpeak sends the plan identifier and purchase token to Dayella’s verification service. The service checks the purchase with Google and acknowledges completed purchases. Stored verification records include a hashed purchase reference, plan, access status, expiry and last verification time. These records support access checks and prevention of duplicate or invalid grants. Subscription verification is separate from recordings, transcripts and handwriting. Contact support below for help with verification records.
Support and beta diagnostics
In-app problem reports and diagnostic files stay on your device until you explicitly share or submit them. Problem reports omit transcript text by default and show any selected inclusion in their preview. The diagnostic file lists app/system versions, device and memory information, selected modes/languages, permission status and recent error codes; it excludes recordings, transcripts, corrections, dictionary and account credentials. Review the preview and remove unrelated personal information before sharing.
The app offers local copy/export and an explicit link to external AI support. Opening that link does not attach your diagnostic file, report text, email, recording or transcript. The external site receives ordinary network information when opened. If you then submit a message, attachment or contact details, Chatbase processes that material under its support/AI service terms; relevant details can pass to Zendesk for a Dayella support ticket. This separate support service is not speech or writing permission. Review its disclosure before sending personal information. The OS share sheet sends a diagnostic file only to the app or destination you choose.
TestFlight feedback and Apple diagnostic reporting can also include app/device details, crash reports and screenshots according to Apple’s reporting flow and your settings. We use submitted reports to respond and diagnose problems, retaining support records as needed to resolve requests and meet applicable obligations. Request access, correction or deletion through our support channel. We do not sell your recordings or transcripts or use them for advertising.
Your choices and deletion
Use the app to delete saved history, meeting/translation recordings, dictionary entries and downloaded models. You can withdraw remote-AI permission, remove API connections, sign out, disable iCloud sync and revoke microphone/Speech Recognition permissions in system settings. Signing out is not account deletion. Keep a copy of anything you need before deleting.
To delete an EnterSpeak managed cloud account, use Delete cloud account in the cloud AI account controls and confirm the loss of any remaining credits. This revokes sessions, removes stored request content/results and payloads, and removes the external Apple or Google sign-in identifier. Internal account, transaction and usage references can remain for accounting, security and prevention of duplicate purchase claims; this is not irreversible anonymisation. Local recordings, iCloud data and exported copies require separate deletion. Without the app, contact support below to request account deletion; we verify ownership and explain records that must be retained. Do not send passwords, identity tokens, API keys or purchase tokens.
iCloud deletion requires enabled, working sync. Removing the app does not itself delete existing iCloud data, original imported files, copies shared with other apps or provider-held data. Manage private iCloud app data in Apple settings where available. Apple and Google subscriptions must be cancelled separately; account deletion, deleting data or uninstalling does not stop renewal.
This website and support chat
Website pages and language links work without an account. The interactive homepage uses preset text and writes to the clipboard only after you select Copy. Cloudflare processes request information needed to serve and protect the site.
The AI support widget is provided by Chatbase. Loading it sends ordinary network request information to Chatbase; messages and any contact details or attachments you submit are processed by Chatbase. The widget may use browser storage to retain chat sessions and preferences. When human follow-up is needed, relevant conversation details and your supplied email are sent to Zendesk to create a Dayella support ticket. Do not share passwords, API keys, payment details or unrelated private text. Website chat does not automatically access your app recordings, transcripts or private iCloud data.
Privacy questions
For support, privacy questions or help with your data, open our Chatbase support chat. If follow-up is needed, you can provide your contact details privately in the conversation.
For technical issues, include your app version, device, operating system version and steps to reproduce the problem. Remove private text from screenshots. Never send passwords, payment details or verification codes.