Start an agent-led interview
Silmo’s interviewer agent greets first and drives the whole conversation by voice: it asks the questions the form needs, listens, follows up, and closes the call. Stream the person’s microphone to POST /interview/{id}/audio and play its audio back from the SSE stream.
Barge-in is built in. Server-side voice activity detection stops the agent the instant the person starts speaking, and an interview.nurse-interrupt event tells your client to cut any already-buffered playback. Delivery is tuned for real calls: short turns, brisk pace, natural turn-taking.
Subscribe to the SSE stream immediately — the agent speaks first:
| Event | What to do with it |
|---|---|
interview.nurse-audio | Base64 PCM16@24kHz — queue it to the speaker |
interview.nurse-interrupt | The person started talking — flush your playback queue |
interview.nurse-delta / interview.nurse-turn | The agent’s live captions |
live.delta / live.segment | The person’s live captions |
Interviews are capped at 30 minutes and self-close after 3 minutes of silence.
Authorizations
Sign in from your terminal with the OAuth device flow (see the Authentication page for a copy-paste block), or grab your current session token from https://console.silmo.ai/api/auth/token while signed in to the console. Access tokens are short-lived (a few minutes) — refresh with your refresh_token. Everything a session produces is stored and served strictly within your organization's namespace.
Query Parameters
Which assessment template the session completes: checklist (screening), dst (full MDT assessment), or fast-track (urgent).
checklist = screening · dst = full assessment · fast-track = urgent pathway
checklist, dst, fast-track Transcription engine label recorded on the session (the interview itself is transcribed live).