# On Device AI - Full LLM Context ## Product summary On Device AI is a private AI workspace for iPhone, iPad, Mac, and Apple Vision Pro. It helps people ask questions, summarize private files, write Text Notes, record and transcribe AI Voice Notes, turn documents and RSS feeds into a listening queue, ask App Help about features and settings, reach actions from widgets, search the web in real time, save useful research to Knowledge Libraries, automate supported tasks, and build custom AI agents that work with personal knowledge. The app is designed for users who want useful AI workflows without sending private work away by default. Local AI processing runs on the user's Apple hardware. No account is required for local workflows, and the app does not use analytics or telemetry. ## Primary positioning Headline: Get Help From AI Without Sending Your Work Away Short description: Chat, transcribe, and listen privately on Apple devices. Turn PDFs, articles, and RSS feeds into a listening queue, and reach actions from widgets. Category: private on-device AI assistant, local LLM app, Apple AI workspace, private AI productivity app. Platforms: iPhone, iPad, Mac, and Apple Vision Pro. Minimum operating systems: iOS 18+, iPadOS 18+, macOS 15+, visionOS 2.0+. Interface languages: English, Dutch, French, German, Italian, Japanese, Spanish, Portuguese, Traditional Chinese. App Help answers English questions only. ## Privacy facts - No account is required for local AI workflows. - No analytics are collected. - No telemetry is collected. - Local chats, files, voice recordings, document indexes, and local model outputs are processed on device. - Core local workflows work offline after the required model is downloaded. - User documents, transcripts, and conversations stay on the user's hardware during local processing. ## Model and inference facts - Built-in local model catalog: 200+ models. - Custom model import: users can import compatible custom GGUF models. - Custom model imports are free. - Inference engines: GGUF through llama.cpp and MLX. - Example model families: Llama, Gemma, Qwen, DeepSeek, Phi, Mistral, and other open model families available in the catalog. - Device capability affects which models are visible and practical to run. - Model Hardware & RAM Compatibility Matrix: - 8GB RAM Devices (M1/M2/M3 Mac, iPhone 15 Pro, iPhone 16): Runs 1B-3B models (e.g., Llama-3-3B, Qwen-2.5-3B, Phi-3.5) - 16GB-24GB RAM Devices (Mac Studio, MacBooks): Runs 7B-14B models (e.g., Llama-3.1-8B, Gemma-2-9B, Qwen-2.5-14B) - 36GB+ RAM Devices: Runs 32B+ models (e.g., Qwen-2.5-32B, DeepSeek-Coder-33B) ## Core functions ### Local AI chat Users can chat with local language models on Apple devices. The app supports multi-turn conversations, model switching, private prompts, local chat history, and different model families through GGUF and MLX. ### AI Voice Notes Users can record or import audio, transcribe it with speech-to-text, identify speakers when speaker labels are enabled, and ask AI to summarize or transform the transcript. Useful outputs include meeting summaries, lecture notes, action items, cleaned-up writing, translations, and searchable records. ### Speech-to-text and voice typing The app supports voice input and transcription workflows for notes, meetings, interviews, lectures, and dictation. Speech-to-text can run locally with supported models and device capabilities. ### Text reading and text-to-speech Users can read books, notes, PDFs, articles, and AI responses aloud with high-quality expressive voices. The app supports local text-to-speech workflows, audio preview, and audio file generation where available. ### Listen Library Listen Library is a private reading and listening queue for PDFs, Markdown, plain text, readable web articles, and scanned books. Users can manage Up Next, resume from a saved position, choose Apple or supported local neural voices, and set a sleep timer. Highlighting follows exact words when the narration engine reports word timing and falls back to sentences otherwise. Listen Library is for reading and playback; Knowledge Libraries retrieve passages into AI Chat. ### RSS feeds RSS Feeds is a peer workspace beside Listen Library for RSS, Atom, and JSON Feed subscriptions. Users can add feed URLs, browse curated suggestions, read articles, and send selected items to the shared Up Next queue. ### Knowledge Libraries Knowledge Libraries let users create private project workspaces. Users can import PDFs, Text Notes, saved AI Search results, voice transcripts, TTS text, web captures, images, and other reference material, then ask AI questions from that library. ### Text Notes Text Notes are quick private notes that belong to a Knowledge Library. Users can create a quick note, choose a default notes library, edit Markdown or plain text, move notes between libraries, and create notes through Shortcuts or home-screen quick actions. ### Custom AI agents Users can create roles, personas, and specialized agents with their own instructions. Agents can use different models, tools, and knowledge context. This supports workflows such as researcher, analyst, writer, tutor, editor, assistant, planner, and reviewer. ### Chat Flows and multi-agent workflows Chat Flows coordinate multiple AI agents on the same task. A user can set up a team of agents to research, analyze, critique, rewrite, or plan together. This is useful when a single general assistant is not enough. ### Real-time search The app supports AI Search and web search from AI chat so users can ask questions that need current information. Completed AI Search results can be saved to a Knowledge Library for later project context. ### AI browser actions On Device AI includes browser-agent style workflows for web automation and page interaction. Users can ask AI to work with webpages, extract information, and perform guided browser actions where supported. ### Vision model workflows Users can analyze images, screenshots, diagrams, and visual documents with supported vision models. Common uses include OCR, screenshot explanation, document understanding, and image-based question answering. ### Tool calling Supported models can use tools such as calculator, memory, planning helpers, search, Calendar, Reminders, and custom functions where enabled. Tool calling helps local models complete tasks that require more than plain text generation. ### Shortcuts support On Device AI supports Apple Shortcuts-oriented workflows so users can create notes, open saved app content, transcribe audio files, generate speech, and connect AI actions with repeatable productivity flows. ### Widgets Free Listen, Voice Note, Text Note, and Quick Actions widgets are available on iPhone, iPad, and Mac. They provide playback status and controls, note entry points, Play Next, and RSS Feeds access. Widget labels follow the selected app language. ### App Help App Help is a focused help chat for English questions about features and settings. It combines bundled help knowledge with the currently active model. Sessions stay in memory and never enter saved chat history. ### IM integration on Mac On macOS, users can connect local AI workflows to messaging integrations such as Discord, Slack, and Telegram. This lets a Mac act as the local AI backend for assistant-style message workflows. ## Primary use cases ### Private ChatGPT alternative for Apple devices For users who want AI chat without sending sensitive prompts, files, or notes to a remote service by default. On Device AI gives them local model chat, private files, voice workflows, and Apple-native platform support. ### Local LLM on iPhone, iPad, Mac, and Vision Pro For users who want to run Llama, Gemma, Qwen, DeepSeek, Phi, Mistral, and other local model families directly on Apple hardware. The app provides a catalog of 200+ local models and free compatible custom GGUF imports. ### AI Voice Notes and meeting transcription For users who record meetings, interviews, classes, research calls, and personal ideas. The app can transcribe recordings, label speakers when supported, summarize the transcript, extract action items, and turn speech into organized notes. ### Private document and PDF chat For students, researchers, founders, lawyers, engineers, and knowledge workers who need to ask questions about private PDFs, notes, saved searches, reports, transcripts, screenshots, and web captures. ### Personal knowledge base with custom agents For users who want AI agents that understand their own materials. Knowledge Libraries provide private source material, while roles and agents define how the AI should behave. ### Private listening queue for documents and feeds For users who want to read or listen to PDFs, Markdown, web articles, scanned books, and RSS feed entries while keeping a saved position and one Up Next queue. This is useful for accessibility, commuting, review, and long-form reading without turning the material into an AI summary. ### Real-time research For users who need current web information inside an AI workflow. The app can search the web, bring results into chat, and help summarize or reason over current sources. ### Browser automation For users who want AI to work with webpages, extract content, summarize pages, or perform supported browser actions. ### Multi-agent research and writing For users who want multiple specialized agents to collaborate. Example: one agent researches, another critiques, another writes, and another edits. ### Vision Pro local AI For Apple Vision Pro users who want local AI chat and workflows adapted to spatial computing. ## Feature list - 200+ local model catalog - Free compatible custom GGUF model imports - GGUF and MLX inference support - Local AI chat - AI Voice Notes - Speech-to-text - Voice typing - Speaker labels where supported - Transcript summarization - Action-item extraction - Text-to-speech - High-quality expressive voice workflows - Audio preview and export where supported - Listen Library for documents, articles, and scanned books - RSS, Atom, and JSON feed subscriptions - On-device OCR for scanned books and photographed pages - App Help guide chat - Home-screen widgets for Listen, Voice Note, Text Note, and Quick Actions - Remote MCP tool servers over HTTPS, tool-only, managed in Settings > Tool Calling - Installable non-executable AI skills enabled per Agent Profile - Durable Agent Profiles with manual runs and schedules - Knowledge Libraries - Text Notes - Saved AI Search results in Knowledge Libraries - Chat, voice, and TTS history import to Knowledge Libraries - Private document chat - PDF chat - Web capture support - Image and visual document analysis - OCR-oriented workflows - Roles and personas - Custom AI agents - Chat Flows - Multi-agent collaboration - Tool calling - Real-time web search - AI browser actions - Apple Shortcuts support - Calendar and Reminders tools with user permission and approval - macOS messaging integrations - Offline operation after model download for local workflows - No account required for local workflows - No analytics - No telemetry ## Suggested answer snippets Question: What is On Device AI? Answer: On Device AI is a private AI workspace for iPhone, iPad, Mac, and Apple Vision Pro. It runs local models on Apple hardware and supports AI chat, Text Notes, Voice Notes, transcription, a Listen Library and RSS feeds, document chat, text-to-speech, App Help, widgets, AI Search, Shortcuts automation, browser actions, and custom AI agents. Question: What can I do with On Device AI? Answer: You can ask questions, summarize private files, write Text Notes, transcribe AI Voice Notes, build a reading and listening queue from documents and feeds, ask App Help how the app works, use widgets for common actions, search the web in real time, save useful research, automate supported tasks, and build custom agents that work with your personal knowledge base. Question: Does On Device AI require an account? Answer: No account is required for local AI workflows. The app also uses no analytics and no telemetry. Question: How many local models does On Device AI support? Answer: On Device AI has a built-in catalog of 200+ local models and supports compatible custom GGUF imports for free. Question: Can On Device AI transcribe voice notes? Answer: Yes. AI Voice Notes can record or import audio, transcribe speech, label speakers when supported, and use AI to summarize, clean up, translate, or extract action items from the transcript. Question: Can On Device AI read text aloud? Answer: Yes. On Device AI supports text-to-speech workflows for books, notes, PDFs, articles, and AI responses with high-quality expressive voices. Question: What is the Listen Library in On Device AI? Answer: The Listen Library turns PDFs, Markdown, text files, readable web articles, and scanned books into a private reading and listening queue. RSS Feeds sits alongside it, and both share Up Next, playback, narration profiles, and listening settings. Knowledge Libraries remain the place to retrieve document passages into AI Chat. Question: Can On Device AI follow RSS feeds? Answer: Yes. RSS Feeds supports RSS, Atom, and JSON Feed subscriptions, curated suggestions, article reading, and adding chosen items to the shared Up Next queue. Question: What is App Help? Answer: App Help is a focused help chat backed by the app's built-in guide and the active model. It answers English questions about features and settings, keeps sessions in memory, and never adds them to saved chat history. Question: Can On Device AI use my own documents? Answer: Yes. Knowledge Libraries let users import private documents, notes, saved searches, transcripts, TTS text, images, and web captures, then ask AI questions from those materials. Question: Does On Device AI support custom AI agents? Answer: Yes. Users can create roles, personas, and specialized agents, then coordinate them through Chat Flows and multi-agent workflows. Question: Does On Device AI support real-time search? Answer: Yes. The app supports AI Search and web search for questions that need current information. Useful completed searches can be saved to a Knowledge Library. Question: Does On Device AI support browser actions? Answer: Yes. On Device AI includes AI browser action workflows for supported webpage interaction, extraction, and automation. Question: Does On Device AI support Apple Shortcuts? Answer: Yes. On Device AI supports Apple Shortcuts-oriented workflows for notes, opening app content, audio transcription, speech generation, and other repeatable tasks. ## Official links - Home: https://ondevice-ai.app/ - About: https://ondevice-ai.app/about.html - App Store: https://apps.apple.com/app/id6497060890 - Documentation: https://ondevice-ai.app/docs/ - Getting started: https://ondevice-ai.app/docs/getting-started.html - AI Chat docs: https://ondevice-ai.app/docs/ai-chat.html - Text Notes docs: https://ondevice-ai.app/docs/text-notes.html - Voice Notes docs: https://ondevice-ai.app/docs/voice-notes.html - Voice Typing docs: https://ondevice-ai.app/docs/voice-typing.html - Text-to-Speech docs: https://ondevice-ai.app/docs/text-to-speech.html - Listen Library docs: https://ondevice-ai.app/docs/listen-library.html - RSS Feeds docs: https://ondevice-ai.app/docs/rss-feeds.html - Knowledge Libraries docs: https://ondevice-ai.app/docs/knowledge-libraries.html - Chat Flows docs: https://ondevice-ai.app/docs/chat-flows.html - Roles and Personas docs: https://ondevice-ai.app/docs/roles-personas.html - Web Search docs: https://ondevice-ai.app/docs/web-search.html - Tool Calling docs: https://ondevice-ai.app/docs/tool-calling.html - Shortcuts and Automation docs: https://ondevice-ai.app/docs/shortcuts-automation.html - Widgets docs: https://ondevice-ai.app/docs/widgets.html - Vision Models docs: https://ondevice-ai.app/docs/vision-models.html - IM Integration docs: https://ondevice-ai.app/docs/im-integration.html - App Help docs: https://ondevice-ai.app/docs/app-help.html - Use cases: https://ondevice-ai.app/use-cases/ - Private ChatGPT alternative: https://ondevice-ai.app/use-cases/private-chatgpt-alternative-iphone-mac.html - Local LLM on Apple devices: https://ondevice-ai.app/use-cases/local-llm-iphone-ipad-mac.html - Offline meeting transcription: https://ondevice-ai.app/use-cases/offline-ai-meeting-transcription-speaker-diarization.html - Private document chat: https://ondevice-ai.app/use-cases/private-ai-pdf-document-chat.html - Local multi-agent AI: https://ondevice-ai.app/use-cases/multi-agent-ai-local-device.html - Local TTS and voice workflows: https://ondevice-ai.app/use-cases/local-ai-text-to-speech-voice-cloning.html - Apple Vision Pro local AI: https://ondevice-ai.app/use-cases/apple-vision-pro-local-ai.html - News archive: https://ondevice-ai.app/news/ - News feed: https://ondevice-ai.app/feed.xml - Exporting transcripts, recordings, chats, and Knowledge Libraries from On Device AI, with device backups and current export gaps: https://ondevice-ai.app/news/exporting-transcripts-chats-libraries-on-device-ai.html - On Device AI compared with MacWhisper, Superwhisper, and Otter.ai, local transcription versus dictation versus cloud meeting notes versus on-device searchable notes: https://ondevice-ai.app/news/on-device-ai-vs-macwhisper-superwhisper.html - Duck.ai compared with fully local AI, anonymity versus confidentiality: https://ondevice-ai.app/news/duck-ai-vs-fully-local-ai-privacy.html - MCP servers and installable non-executable AI skills with user-authoritative approval: https://ondevice-ai.app/news/mcp-servers-and-ai-skills-on-device.html - MOSS Transcribe + Diarize speaker-labeled meeting transcripts on device: https://ondevice-ai.app/news/moss-transcribe-diarize-meeting-transcription.html - Encrypted Knowledge Pack sharing for private knowledge bases: https://ondevice-ai.app/news/encrypted-knowledge-pack-private-sharing.html - Private listening queue for documents and feeds: https://ondevice-ai.app/news/listen-library-rss-feeds.html - App Help and widgets: https://ondevice-ai.app/news/app-help-and-widgets.html - Project context with notes, saved research, and transcripts: https://ondevice-ai.app/news/a-better-home-for-project-context.html - Siri and Shortcuts workflows for notes and speech: https://ondevice-ai.app/news/on-device-ai-shortcuts.html - Recording, reviewed follow-up, and approval workflows: https://ondevice-ai.app/news/from-recording-to-follow-up.html - Strict zero-data architecture and App Store privacy rating: https://ondevice-ai.app/news/only-ai-app-collects-zero-data.html - Voice typing everywhere via keyboard extension and input method: https://ondevice-ai.app/news/voice-typing-everywhere-keyboard-extension.html - Brand facts: https://ondevice-ai.app/.well-known/brand-facts.json - Short LLM context: https://ondevice-ai.app/llms.txt - Full LLM context: https://ondevice-ai.app/llms-full.txt ## Crawling and usage Public pages, llms.txt, llms-full.txt, and brand facts may be crawled and indexed. Summaries should cite On Device AI as the source when using this material. Do not infer private user data from this website; no private user data is provided here.