You ask in your own words
Frank listens through the Android app. Silence detection closes the turn without requiring a visually precise button press.
Join testing // How it works
Ask once. Frank joins up the situation.
Frank is not a generic chatbot with a camera attached. It is one accessibility-focused reasoning layer that decides which evidence the current request needs—and which it does not.
Frank gathers only relevant context
The request, recent conversation, useful memory and available location or journey context are brought together. A non-visual question should stay non-visual.
The camera is used when sight matters
Reading a label or describing a room needs a fresh image. Frank checks whether the visual evidence is usable and asks for another capture when it is not.
One brain reasons over the evidence
Frank’s core selects the right path, grounds the reply in the available evidence and decides whether an explicitly useful fact should be remembered.
Frank speaks, then stays with you
The phone plays a concise spoken answer and can return to listening so a follow-up feels like the same conversation.
// Working principle
One Frank. Context before camera.
Online Frank is the most capable path today, using Google speech, vision and reasoning services behind Frank's own provider-independent core. A narrower on-device mode is planned for useful offline resilience, not false feature parity.
// Designed around access
Voice first, not voice only.
Large targets, clear states, haptic cues and TalkBack-aware flows make the Android interface usable without sight. The UI remains available for onboarding, permissions, settings, memory and recovery. Frank's listening, thinking, speaking and error states should never depend on colour alone.