Most voice-notes apps run server-side transcription and store your audio in the cloud. ThoughtSift is built the other way around.
Screen-off capture
A volume-button or voice trigger starts recording with the phone locked — no unlock, no app navigation, no broken walk. A haptic tap confirms when capture completes. The capture habit fits your day instead of interrupting it.
Classified locally
ThoughtSift classifies your thought on-device via Core ML. When transcription runs on-device, your voice audio never leaves the phone. You see a confidence score, and a one-tap correction trains the model — so it gets better without sending your recordings anywhere.
Export-and-evaporate
After you export a capture to Notion or Obsidian, a single tap deletes both the transcript and the cached audio from the device. It’s a privacy posture that cloud-only tools — which retain audio on their servers — structurally can’t match.
Privacy isn’t a setting bolted on at the end here. It’s the architecture.