Last updated: July 17, 2026
QalbeQal is an Amharic-first voice-to-text service. Audio has to leave your device to be transcribed. Ordinary dictation is not saved by QalbeQal as a training recording. Training sharing is a separate choice, off by default.
We do not sell personal data, use it for third-party advertising, or track you across unrelated apps and websites.
Recording begins only after you start it. Your audio is sent through QalbeQal's cloud service to speech-processing providers so a transcript can be returned. If optional text cleanup is enabled, transcript text and relevant personal-dictionary terms may also be processed to repair spelling, spacing, or punctuation.
Standard dictation audio is used to complete the request and is not added to QalbeQal's training dataset unless you separately opt in. QalbeQal does not intend text cleanup to translate, summarize, or change the meaning of your words.
The iPhone and Mac apps may keep transcript history, preferences, personal-dictionary terms, and learned corrections on your device. Local history contains text and timing information, not the original recording. You can edit or clear this information in the app.
To operate limits, prevent abuse, understand reliability, and show usage impact, QalbeQal records limited metadata such as a random device identifier or account identifier, platform, request time, audio duration, transcript character count, success or error status, processing time, and retry count. These records do not contain transcript text or the audio recording.
QalbeQal may turn that metadata into collective statistics such as total voice sessions, estimated words dictated, audio time, and estimated typing time saved. The founder dashboard receives aggregate totals—not individual transcript text, recordings, names, email addresses, or a list of identifiable users.
You can use some parts of QalbeQal without creating a named account. If you create one, we store the account details needed to sign you in, such as your email address and authentication record. A random device identifier may still be used for rate limits and service protection.
If you turn on training sharing, future dictations may contribute the audio and associated transcript to a private dataset used to improve Amharic speech-to-text (STT). Nothing recorded before you opted in is added automatically. Turning sharing off stops future contributions.
Training contributions remain available for model-development work until deleted under our retention needs or after a valid deletion request. Removing a contribution from storage cannot undo learning already incorporated into a model that was trained before the deletion.
When you use the Telegram bot, Telegram sends QalbeQal the voice note and basic account information needed to receive the message and reply. QalbeQal may keep the Telegram account ID, language and access settings, quota records, service metadata, and feedback needed to provide and protect the bot. Telegram has its own privacy practices.
Some experimental providers or features are available only after you turn them on. When you do, the audio or text needed for that feature is sent to the named provider. The app will identify an optional provider before you enable it when its data practices differ meaningfully from the standard service.
QalbeQal uses service providers for cloud infrastructure, authentication, speech processing, optional language processing, website hosting, and message delivery. Current providers include Supabase, Google, ElevenLabs, Vercel, and Telegram where their service is used. They receive only the information needed for their role and process it under their own terms and QalbeQal's configuration.
Local information remains until you clear it or remove the app. Account and service metadata is kept while needed to provide the service, enforce limits, investigate abuse, meet legal obligations, and understand reliability. Optional training contributions are handled as described above.
We use access controls and private server credentials to protect cloud data. No internet service can promise perfect security, so do not dictate information you would not be comfortable sending to a cloud transcription service.
You can clear local history, remove dictionary terms, turn text cleanup off, sign out, delete your account where available, and turn training sharing off. You can ask us to delete server-side information reasonably linked to your account, device identifier, Telegram ID, or training contributions.
Both deletions are available directly in the app: delete your account under Settings → Account, and delete device-linked data (contributed clips, ratings, older usage records) under Settings → Privacy → Details & my data — each takes effect immediately. You can also send a deletion request through the QalbeQal Telegram bot: @QalbeQalbot, or by email to hello@yaredstudio.com. We may ask for enough information to verify that the records belong to you.
QalbeQal is not directed to children under 13. If you believe a child provided personal information, contact us so we can review and delete it where appropriate.
We may update this policy as the product changes. We will change the date above when we do.
Questions, support requests, and deletion requests can be sent through the QalbeQal Telegram bot: @QalbeQalbot, or by email to hello@yaredstudio.com.