agent-voice-dictation

Lean voice dictation primitives for agent composer UIs.

V1 intentionally stays small:

• browser recorder + audio levels
• direct STT clients for AssemblyAI, Deepgram, OpenAI, Gladia, and ElevenLabs
• OpenRouter transcript polish client
• a thin pipeline that composes speech detection and polish

Host apps own secrets, UI, persistence, and insertion into their composer.

Boundary

speech/ talks to speech-to-text APIs only.

openrouter/ talks to OpenRouter only.

recorder/ captures browser microphone audio only.

pipeline/ composes those pieces.