This archive covers product updates, workflow improvements, and product thinking behind private AI experiences that run directly on your device.
Duck.ai strips your identity and forwards the prompt to OpenAI or Anthropic. A local model never sends it. An honest comparison of two real privacy designs, including where local loses.
Read article →Remote MCP tools and imported skill packages extend what your local agents can do. No subprocess launching, no executable hooks, no self-granted approvals, and credentials that never leave the Keychain.
Read article →The new MOSS model returns transcript text and speaker turns from the same on-device pass, handles long meetings in bounded passes, and works for recordings, imports, re-transcription, Mac Voice Typing, and Shortcuts.
Read article →Export a Knowledge Library as an encrypted Knowledge Pack, protect it with a password, and hand it to a colleague. The recipient imports a library that is searchable immediately, with no cloud account and no re-indexing.
Read article →Gather PDFs, articles, scanned books, and RSS feeds into one private queue you can read or listen to, with Up Next, a sleep timer, and on-device OCR.
Read article →Ask how a feature works without mixing help into saved chats, then reach listening and note actions from widgets on iPhone, iPad, and Mac.
Read article →Keep notes, saved research, transcripts, and project files together in a private Knowledge Library, then use that context when you return to the work.
Read article →Use Siri and Shortcuts to create notes, open saved work, transcribe audio, and generate speech for the tasks you repeat most.
Read article →Review a private transcript, keep the context that matters, and approve a Calendar or Reminder change only when it looks right.
Read article →Start from a focused workflow: local LLMs, private document chat, offline meeting transcription, local TTS, multi-agent AI, or Vision Pro.
Explore use cases →Most "private" AI apps still collect analytics or route queries. See how On Device AI maintains a strict zero-data architecture and a "Data Not Collected" rating.
Read article →On Device AI records, transcribes, labels speakers, and summarizes voice notes offline with Apple STT, Whisper, Parakeet, Nemotron, and Qwen3-ASR model choices.
Read article →On Device AI adds more local TTS choices, saved cloned voices for supported models, and optional noisy-speech cleanup for user-provided reference audio.
Read article →On Device AI's Pro speaker diarization turns Voice Note recordings into private meeting transcripts with speaker labels, advanced tuning controls, and export to text, SRT, or Markdown.
Read article →Our custom iOS keyboard extension and macOS input method bring secure, offline AI-powered voice typing with real-time Whisper transcription to any app.
Read article →On Device AI's Browser Agent runs a local ReAct loop inside a real WebKit session, so your private model can navigate sites, fill forms, and extract data without handing the page to a hosted browser farm.
Read article →On Device AI adds two multi-agent modes: subagent delegation, where a main agent routes tasks to specialized experts, and Chat Flow for parallel agent responses. Both run locally on iPhone, iPad, and Mac.
Read article →Recent supply chain attacks on major web libraries show the fragility of modern app development. Here is why On Device AI is built directly on Apple's native technologies to keep your data secure and entirely local.
Read article →On Device AI for macOS adds IM Communication—connect Discord, Slack, or Telegram bots to a private, on-device AI assistant that processes messages locally without cloud processing.
Read article →Apple's March 2026 MacBook lineup introduced three genuinely different machines—Neo, Air M5, and Pro M5 Max. Here's what each tier delivers for on-device AI workflows with On Device AI.
Read article →This editorial explains what the best local private AI apps should include in 2026, from MLX and GGUF support to voice, vision, knowledge retrieval, and Apple-device workflows.
Read article →On Device AI turns scattered research inputs into a private Knowledge Library workflow, making it easier to retrieve grounded context from PDFs, web captures, OCR images, and notes.
Read article →