🎙️ Talk to Bobby & the Advisor Board

loading…

Who Mode

idle

Advisor Voice (everyday, cheap): hold-to-talk → AssemblyAI speech-to-text (the engine configured in Bobby's UI; browser SpeechRecognition is the fallback) → LiteLLM (Gemini 2.5 Pro) with the advisor's persona → ElevenLabs voice. Bobby Live / Translate (premium, opt-in): your browser connects straight to OpenAI's Realtime API with a short-lived token. Off by default — start the gateway with RV_ALLOW_OPENAI_REALTIME=1 (npm run voice:openai) when you want them. gpt-realtime-2 is the priciest mode; use for the conversation, then disconnect.

Persona prompts: agents/advisor-board/<slug>/SYSTEM.md · voices: skills/advisor-voice/voices.json · gateway: services/realtime-voice/server.mjs. Research/personal tool — advisor outputs are not professional advice.