Zero network traffic in default configuration
Screen capture (`ScreenCaptureKit`), audio encoding (`AVAssetWriter`), speech recognition (`SpeechAnalyzer`), and default AI (`FoundationModels`) run 100% locally on your Mac.
Metis Meet treats every transcript, audio track, and indexed source file as sensitive local data.
Screen capture (`ScreenCaptureKit`), audio encoding (`AVAssetWriter`), speech recognition (`SpeechAnalyzer`), and default AI (`FoundationModels`) run 100% locally on your Mac.
When using the Cursor CLI backend, every invocation runs under `/usr/bin/sandbox-exec` denying `file-read*` and `file-write*` on `$HOME` and `$TMPDIR`, paired with a throwaway run folder, a deny-all `cli-config.json`, and a pre-flight canary test.
`SecretFilter` strips `sk-*`, `AKIA*`, `ghp_*`, `xoxb-*`, PEM private keys, JWTs, and `password`/`secret` assignments before chunks are written to SQLite or sent to any AI model.
Cloud API keys are stored in the macOS login Keychain under `com.aliabbasi.MeetingRecorder.ai` with `kSecAttrAccessibleWhenUnlocked`. Sending transcript text to any remote host requires a one-time user consent confirmation.