Voice intake — MVP

Talks to OpenAI Realtime with our HM intake voice prompt. Mic + audio runs browser-direct; the full API key stays on the server. Transcript appears live as you both speak.

Job description (optional)

Paste the JD if you have one. The agent will use it as context and skip re-asking what's already in it. Leave blank to build the spec from scratch through conversation.

Status: idle

Live transcript

Start the intake to see the transcript appear here.

MVP notes: this is Path B from the audio prep doc — OpenAI Realtime drives the conversation end-to-end with our HM voice prompt as system message. No tool calls (no loadHMContext / submitJobSpec) yet. Output is the transcript only. Phase 2 will route turns through our existing Claude agent loop so we keep the structured-output discipline.