/sessions endpoint

wss://api.medera.info/sessions is the stateful, diarized live-capture WebSocket for therapy sessions. Same engine, same calibration, same structured output as /transcribe and /transcripts — but architected for the full 50-minute encounter with 2-speaker diarization, in-line C-SSRS markers, and structured instrument-score emission.

Connection

wss://api.medera.info/sessions?session_id=<id>&token=<jwt>

Configuration (client → server)

{ "type": "configure", "language": "en-US", "diarization": true, "speakers": 2, "interim_results": true, "instruments": ["phq-9", "gad-7", "audit-c", "c-ssrs"], "vocabulary": { "formulary": ["sertraline", "escitalopram", "lamotrigine"], "programs": ["IOP", "PHP", "PHP-step-down"], "clinicians": ["Dr. Khalil", "Dr. Patel"] }, "features": ["soap", "codes", "span_anchors", "crisis_markers"]
}

Events (server → client)

Event	Schema
`transcript.partial`	`{ text, speaker, start_ms, end_ms }`
`transcript.final`	`{ text, speaker: "clinician" \| "patient", start_ms, end_ms, span_id }`
`instrument.item_scored`	`{ instrument, item, score, transcript_anchor }`
`instrument.completed`	`{ instrument, total_score, severity, items, anchors }`
`crisis.detected`	`{ band, criteria, transcript_anchor, escalation_id }` (C-SSRS)
`note.section.update`	`{ section, text, anchors: [span_id] }`
`code.suggestion`	`{ system, code, display, anchor }` (ICD-10, CPT)
`audio.event`	`{ kind: "silence" \| "clipping" \| "low_snr" }`
`error`	`{ code, message, recoverable }`

Lifecycle

client opens WS → server: { connected, session_id } → client: { configure,... } → server: { configured } → client: <PCM binary frames>... → server: { transcript.partial }, { transcript.final } → server: { instrument.item_scored } (continuous) → server: { instrument.completed } → server: { crisis.detected } ← never gated → server: { note.section.update } → server: { code.suggestion } → client: { close } → server: { finalized, session_id }

Audio

16 kHz 16-bit PCM mono recommended
Opus at 32 kbps for bandwidth-constrained edges
20 ms frames (320 samples at 16 kHz)
See Audio Configuration for full formats

What you get back at session close

{ "session_id": "ses_abc", "transcript": { "segments": [...], "spans": [...] }, "diarization": { "speakers": 2, "speaker_time_ms": { "clinician": 1820000, "patient": 720000 } }, "instruments": { "phq-9": { "total": 14, "severity": "moderate", "items": [...] }, "gad-7": { "total": 11, "severity": "moderate" }, "c-ssrs": { "risk_band": "low", "items": [...] } }, "codes": { "icd10": [...], "cpt": [...] }, "notes": { "soap": {...} }, "crisis_events": []
}

C-SSRS crisis markers are non-bypassable. When crisis.detected fires, the practice escalation tree is invoked immediately. The agent cannot gate, suppress, or delay the event.

What’s next

/transcribe

Stateless between-visit dictation.

/transcripts

Async batch processing.

Diarization

2-speaker separation.

Audio configuration

Formats and frame size.

​Connection

​Configuration (client → server)

​Events (server → client)

​Lifecycle

​Audio

​What you get back at session close

​What’s next

/transcribe

/transcripts

Diarization

Audio configuration

Connection

Configuration (client → server)

Events (server → client)

Lifecycle

Audio

What you get back at session close

What’s next