Mmesut.uk
Home Work Apps About Contact
← OnDevice Classifier

OnDevice Classifier Privacy Policy

OnDevice Classifier supports local inference and optional remote text classification. This policy explains which mode keeps your input on your device and which services receive data when you select a remote connection.

Effective: October 12, 2026

At a Glance

  • Downloaded-model inference runs on your device
  • Optional API connections send input text and task definitions to your selected provider
  • Image classification is local in this version
  • No developer-operated analytics or user-account service

1. Local models and inputs

Compatible downloaded models process your text or image on your device. The developer does not receive the contents of local classification runs. Model files are stored in the app’s local storage.

2. Optional model API connections

When you choose an OpenAI, Anthropic, Gemini or custom OpenAI-compatible connection, the app sends the input text, category definitions and task instructions directly to that endpoint. The API key identifies your provider account. The provider may retain requests, responses and standard network metadata under its terms and your account settings. OnDevice Classifier does not enforce zero retention across those services. Review your provider’s policies before sending personal or confidential content. Provider practices differ by account tier and region. Google’s unpaid Gemini API tier may use inputs and outputs to improve products and machine-learning technology, and human reviewers may review that content. Refer to the provider’s current terms and data controls; the app does not train its own models on your input.

3. Apple Private Cloud Compute

When you select Apple Private Cloud Compute, your text request is processed through Apple’s supported cloud model service. Availability depends on your device, OS and Apple’s service conditions. Apple’s privacy and Private Cloud Compute protections govern that processing.

4. Model downloads and Hugging Face

Downloading models contacts Hugging Face and its file-delivery hosts for the selected repository, revision and files. These services receive standard network request information. An optional Hugging Face token is sent to Hugging Face to access repositories your account can download; the app strips that credential from cross-host CDN redirects.

5. Credentials and saved history

Provider API keys and the optional Hugging Face token are stored in the device’s Keychain. They are not included in model manifests or run-history exports. Saved text runs can include their input text, categories, result and model details. Original image or audio bytes are not saved in run history.

6. Your controls

Choose a local model to keep classification input on your device. You can remove saved API connections and tokens, delete downloaded models, and delete history in the app. Provider-side retention and deletion are managed with that provider.

7. Tracking and contact

The app has no advertising, tracking SDK, developer-operated analytics or separate user-account service. If you contact support by email, the information you send is used to respond to your request. For privacy questions, contact mesutcydev@gmail.com. This policy covers the app; the mesut.uk website has its own site privacy policy.

Support Back to OnDevice Classifier
Work Privacy Support Contact

© 2026 Mesutcy. All rights reserved.