Enclave AI

Enclave AI runs downloaded language models on iPhone, iPad and Mac for text chat, voice conversations and questions about documents. Local processing works offline after the app and models are downloaded. The current App Store listing requires iOS or iPadOS 27, or macOS 27 on an Apple silicon Mac.

The app offers models from families including Llama, Qwen, SmolLM and Gemma. Mobile and Mac model libraries differ to suit each device's resources. PDFs and text files can be added to conversations for summaries and questions. Custom assistants save prompts and instructions for different tasks.

Voice input uses Apple's on-device speech recognition, and spoken replies use on-device speech synthesis. With a local model selected, the answer is generated on the device too. Siri and Apple Shortcuts can invoke a chosen local or cloud model.

Conversation history is stored locally. Optional cloud requests pass through Enclave's servers to OpenRouter and the selected model provider. These requests can include relevant conversation history, attachments and transcribed speech. The interface indicates when data leaves the device; cloud processing requires a connection and follows the provider's data policies.

Local prompts and answers are not sent to an AI service for processing. The app also uses Firebase Analytics for usage events and Crashlytics for crash diagnostics, so local inference does not mean that the app makes no network requests or collects no diagnostic data.