Enclave AI

Enclave AI is an AI chat bot tool. Enclave AI runs open source language models entirely on iPhone and Mac, with offline chat, local voice input, document Q&A, custom assistants and Shortcuts automation. No account and no cloud by default.

Enclave AI is a chatbot for iPhone and macOS that runs open source language models directly on the device. There is no server round trip and no account registration: the model file is downloaded once, then every prompt, reply and voice interaction is processed locally. The pitch is straightforward. Anything typed into Enclave stays on hardware the user already controls, which makes it usable for material that would be awkward to paste into a cloud assistant, from client documents to medical notes to unpublished drafts. In daily use the app looks like most chat interfaces, with a model manager layered on top. On iOS the picker covers mobile tuned options such as Llama, Qwen, SmolLM and Gemma, plus distilled DeepSeek R1 builds; the macOS version can load larger models because it has more compute to draw on. Users can define custom assistants with their own system prompts, personalities and areas of expertise, and each assistant keeps a separate conversation thread. PDFs and plain text files can be attached to a chat and queried in place. Voice input leans on Apple's on-device speech recognition and synthesis, so speaking to the assistant never transmits audio. Conversation history is stored encrypted on the device, with export and automatic deletion controls. The Shortcuts and Siri integration is where Enclave becomes more than a chat window. AI actions can be dropped into automations and each shortcut can be pinned to a specific model, so a sensitive summarization task might run locally while a harder reasoning job is pointed at a cloud model through the optional OpenRouter connection. That OpenRouter hook is the one deliberate exception to the offline design, and the app marks when processing leaves the device. The realistic buyer is an Apple user with a privacy constraint and modest expectations about raw model quality. Small local models handle brainstorming, rewriting, summarization and document Q&A well; they are weaker on broad general knowledge and current events. Anyone who needs frontier level reasoning most of the day will find a local model limiting, and should treat Enclave as a private second opinion rather than a full replacement for cloud tools. It also stays inside Apple's ecosystem, so Windows and Android users have no path in.