Artfull AI

Draft · 13 October 2026

Privacy

This is a review draft, not a published privacy policy. It describes the current 0.0.1 development design and must be checked against the final app, services, hosting and App Store privacy answers before release.

Who is responsible

Artfull AI is developed by Joe Woods. The intended privacy and support contact is support@artfull-ai.com; the mailbox is not yet verified.

What the app processes

When you start a conversation, Artfull AI processes the prompt and any images or extracted document text you choose to attach. Depending on your selected connection, that material is either processed on your device or transferred to the server or AI provider you configured. The app also handles connection details, model selections, chat titles and response metadata needed to operate its features.

Where conversations go

Before a public release, the app still needs a dedicated, informed consent flow for third-party AI transfer. You should not use the development build for sensitive material unless you understand and trust the selected destination.

History and imported files

Direct-server and on-device conversations are kept separate by connection. Local histories, imported documents and partial model downloads use iOS file protection. Basic is planned to retain only the current conversation context; saved and reopened history is planned as an Advanced capability. Existing stored history must not be deleted merely because access changes.

Credentials

Router sessions and provider API keys are stored as connection-scoped items in the Apple Keychain using device-only, when-unlocked protection. Direct Ollama and LM Studio requests do not receive Custom LLM Router credentials. The app must surface credential failures rather than treating a missing value as valid.

Statistics

The local statistics journal records response metadata such as provider, model, timing and token counts when available. It is designed not to store prompt text, response text or imported document contents. Missing measurements are shown as unavailable rather than as zero.

Model catalogue and downloads

When you browse or download supported MLX models, the app may contact Hugging Face services and content-delivery hosts to retrieve public catalogue metadata and model files. Those services can receive standard network information such as your IP address and request details. Their own policies apply. Downloads are user-initiated, bounded and pinned to a model revision; licensing and full real-download validation remain release checks.

Device permissions

Photos, camera, files, microphone, speech recognition and local-network access are requested only for the related feature. Selected documents and images are processed for import and may be sent with a request to the selected service. Speech availability and system behaviour can vary by language, device and operating-system configuration.

The website

This review website contains no analytics, advertising pixels or third-party tracking scripts. The eventual hosting provider may keep ordinary access logs such as IP address, browser information, requested pages and timestamps for security and operation. Hosting and retention have not yet been selected.

Third-party policies and control

Artfull AI does not control the retention, training, account or deletion practices of servers and providers you choose. Review their policies and configure them appropriately. You can remove local conversations and downloaded models through app controls as those features are completed; Router-held history must be managed through that service.

Security and limits

Artfull AI uses scoped credentials, iOS file protection, bounded downloads and cross-origin redirect checks. These engineering measures are not a security certification. Plain HTTP on a local network is unencrypted, and broader hostile-input, provider, background and device testing remains open.

Changes before release

This draft will change as purchase controls, consent, deletion and retention controls, hosting and final provider behaviour are completed. A dated, verified policy and working contact must be published before any App Store submission.