Voice Agent: Talk to Oscar
/voice is protected by sign-in, not a route-level paid-tier gate. It requires microphone permission and working audio/network support.
Date: 2026-09-30 Purpose and access /voice is protected by sign-in, not a route-level paid-tier gate. It requires microphone permission and working audio/network support. Availability is configuration-dependent: speakerall uses dictation for everyone; adminonly permits the live agent only for admins; liveall permits all users. A configured provider is also required. These source modes do not verify the current deployment. Workflow Open /voice and wait for voice configuration/account loading. In the rendered console, start the microphone/session and grant the browser permission. Speak a question. Live mode connects to the hosted agent; dictation records/transcribes, asks Oscar and reads the result. Wait through connection/research/speech activity; review the transcript rather than assuming unheard audio is an empty answer. Stop/end the session when finished. Use /listen for the separate playback-first experience. Results, limits and states Live transcript synchronization supports conversation history; failed persistence or an unfinished session is not guaranteed saved history. Connection/token, transcription, research and playback can fail independently. Follow the console error, check microphone/site permission and retry; an empty transcript before speaking is normal. Language selection requests the corresponding agent; absent configuration falls back to English, so Korean-specific audio is not promised. Live voice is a third-party provider boundary, not a replica of all interactive chat cards or a trading confirmation channel. Use the actual form/card for consequential approvals. Admin broadcast controls are admin-only, not a subscriber benefit. No native-app or deployment parity is claimed.