> ## Documentation Index
> Fetch the complete documentation index at: https://docs.medera.info/llms.txt
> Use this file to discover all available pages before exploring further.
# Co-Therapy Agent
> Silent multimodal in-session clinical intelligence.
**Category:** Clinical Operations · **Channels:** Vocal acoustics · Facial physiology · Linguistic content · Assessments
The Co-Therapy Agent is a silent multimodal partner — it **never** speaks to the patient. It runs the [Vocal Acoustic Engine](/multimodal/vocal/overview) and [Facial Physiological Engine](/multimodal/facial/overview), fuses outputs through the [Neurobehavioral Construct Computer](/multimodal/rdoc/overview) into 15 named construct activations, and surfaces the resulting canonical payload to the clinician on a private channel.
## Pipeline
## Endpoints
| Method | Path | Purpose |
| :----- | :------------------------------------------- | :-------------------------- |
| `POST` | `/api/co-therapy-sessions` | Create session (backend) |
| `POST` | `/api/co-therapy-sessions/{id}/start` | Start session + invitations |
| `POST` | `/api/co-therapy-sessions/{id}/join` | Record participant joining |
| `POST` | `/api/co-therapy-sessions/{id}/leave` | Record participant leaving |
| `POST` | `/api/co-therapy-sessions/{id}/transition` | Manually transition state |
| `POST` | `/api/multimodal-therapy/analyze-session` | Full multimodal analysis |
| `POST` | `/api/multimodal-therapy/analyze-audio-only` | Audio-only analysis |
## Configuration
| Field | Description |
| :------------------------------- | :-------------------------------------------------------- |
| `audio_enabled`, `video_enabled` | Channel toggles |
| `rdoc_constructs` | Optional subset of the 15 named constructs |
| `clinical_context` | Chief complaint, active diagnoses, recent screener scores |
| `consent_recorded` | Per-session consent flag (required) |
## Composable experts
| Expert | Role |
| :----------------------------------- | :----------------------------------------------------------- |
| **Vocal Prosody Expert** | F0, jitter, shimmer, HNR, speaking rate, prosodic flatness |
| **Affect Dynamics Expert** | Facial affect intensity, valence, congruence |
| **Neurobehavioral Construct Expert** | Neurobehavioral Construct activation, severity, contributors |
| **Linguistic Content Expert** | Topic, valence, certainty, pronoun shift |
## Quality gates
* **Audio**: ≥ 1.0 s, valid `voiced_fraction`
* **Video**: ≥ 30 s for HR, ≥ 60 s for HRV / BP, SNR > 3.0
* **Analyzer**: minimum 20 words of transcript + 30 s of session signal On quality failure, the engine returns `confidence: 0.0` and the analyzer falls back to a Medera reasoning transcript-inferred analysis marked `requires_human_review: true` — **never fabricated values**.
## Compliance guarantees
* The Co-Therapy Agent **never** addresses or interacts with the patient. - Recording requires explicit per-session consent (`consent_recorded: true`). - Construct activations report `contributors[]` so clinicians can audit feature-level influence.
***
## Related
Run a multimodal session.
Engine deep dive.
15 named constructs.
REST surface.