Ask me anything

I stream answers as I think, and speak them out loud as I write.

Settings

Generation

Generate URL LLM Model LLM Base URL Enable RAG RAG Index Name

Text-to-Speech

Enable TTS for assistant replies Hide assistant text · show only audio Advanced streaming · pipeline TTS while LLM streams TTS chunk size (words) Tokens are buffered until this many words OR a sentence punctuation mark is hit, whichever comes first. TTS URL TTS Model Speaker ID Sample Rate (Hz)

Speech-to-Text

Hide transcription · show recording in chat ASR URL Language Manual recording · tap mic to start/stop, one clip to ASR When on, the mic does not use streaming VAD or wake word—only audio between start and stop is sent to ASR. Turn off to use Silero VAD and optional wake word below. Streaming VAD · Silero segments + ASR each utterance End-of-speech silence (seconds) Server-side silence duration before a phrase is sent to ASR (WebSocket pipeline). Min. speech before ASR (seconds) Shorter segments are discarded as noise; default 2s. Wake word (streaming VAD) Web Speech listens for this word first (e.g. “hi”). Interrupts reply when heard; then records your command. Leave empty for always-on VAD (no wake). After wake · end-of-command silence (seconds) After wake · min. command speech (seconds)