Shadow starts from a complaint many knowledge workers share about meeting assistants: the bot that joins your call, the note-taker that locks your history behind a subscription, and the data that leaves your machine for someone else's servers. Shadow's answer is a native Mac app that operates entirely in the background. It has no bot that joins the meeting, auto-detects when a call begins, records and transcribes it locally, and stops on its own when the call ends. Speaker identification is done in real time, so transcripts arrive labelled with who said what.
The output is deliberately portable. Every meeting is written to your Mac as Markdown — transcripts, notes, and Skill results — with every call and every person linked into a knowledge base. You can open the folder as an Obsidian vault, sync it with iCloud Drive or Google Drive, search it with Spotlight, or point any AI agent at it. Shadow emphasises that there is a built-in advantage: because agents read the raw files rather than a connector's filtered view, tools like Claude, Codex, or OpenClaw get the complete context.
Between calls, Shadow is more than a note-taker. Press a keyboard shortcut on any screen and it can see what you're looking at, hear what you say, or read selected text, then run an Action Skill to produce an email, a set of notes, or code. Skills are just prompts you can edit or write from scratch, and a webhook can forward results to Zapier, n8n, or Make. The privacy stance is equally concrete: transcription is local, audio does not leave the device, meeting content is never used for model training, and you choose which Skills are allowed to call AI at all.
