Viora

A conversational mental-health signal pipeline. You send us turns; we return continuous signal scores, structured imminence fields, and an actionable R-level.

10 continuous signals

Self-harm, suicidal ideation, anxiety, depression, substance use, eating disorder, trauma, psychosis, burnout, crisis severity. Each in [0, 1].

Structured imminence

Per-user-turn intent / plan / means / timeframe plus emotional state, protective factors, co-signals. Catches what continuous scores miss.

R-level state machine

R0 → R1-mid → R1-high → R2 with sticky-floor. A claim of 'I'm fine' without structured evidence doesn't auto-downgrade an active crisis.

How it fits in your stack

POST /v1/sessions/{id}/messages
Authorization: Bearer vrk_...
Content-Type: application/json

{ "role": "user", "content": "I haven't slept in days" }

→ 200 OK
{
  "prs": 0.62,
  "scores": { "depression": 0.62, "anxiety": 0.51, ... },
  "r_level": "R1-high",
  "session_r_level": "R1-high",
  "forced_by_imminence": false
}

Sub-second per turn. Local models, no LLM dependency on the scoring path. See the API quickstart.