Recall

AI companion for dementia patients in Sri Lanka. Ama speaks with patients using their memory book; caregivers receive session summaries after each call.

Run locally (two terminals)

Terminal 1 — Python voice agent (LiveKit + Beyond Presence):

cd livekit-bey-agent
cp .env.example .env
# Fill in LIVEKIT_*, GOOGLE_API_KEY, GOOGLE_APPLICATION_CREDENTIALS, BEY_*
uv run agent.py dev

Terminal 2 — Next.js frontend:

npm install
cp .env.local.example .env.local
# Fill in LIVEKIT_* (same project as agent)
npm run dev

Open http://localhost:3000.

LiveKit dispatch rule

In LiveKit Cloud → Agents → Dispatch rules:

Field	Value
Rule type	Individual
Room prefix	recall-room
Agent	ama

Environment variables

Next.js (.env.local)

Variable	Purpose
LIVEKIT_URL	Server-side LiveKit URL (wss://...)
LIVEKIT_API_KEY	API key
LIVEKIT_API_SECRET	API secret
NEXT_PUBLIC_LIVEKIT_URL	Client connect URL (usually same as LIVEKIT_URL)
SESSION_WEBHOOK_SECRET	Bearer token for POST /api/sessions/end (same as agent .env)
USE_SUPABASE	false = in-memory sessions (no DB). Auto-on if Supabase URL + service role key are set

Run without Supabase (local demo)

1. Set USE_SUPABASE=false in .env.local (or leave out Supabase keys).
2. Set SESSION_WEBHOOK_SECRET and the same secret in livekit-bey-agent/.env.
3. Set RECALL_APP_URL=http://localhost:3000 in the agent .env.
4. Start Next.js + agent as usual. Transcripts persist until you restart npm run dev.

For production persistence, use Supabase: run supabase/migrations/001_sessions.sql, set USE_SUPABASE=true, and add NEXT_PUBLIC_SUPABASE_URL + SUPABASE_SERVICE_ROLE_KEY.

Python agent (livekit-bey-agent/.env)

See livekit-bey-agent/.env.example. Do not wrap values in quotes.

Routes

Route	Description
/	Marketing landing page
/patient	Patient home
/patient/start	Handoff — memory chips, begin session
/patient/loading	Token fetch + agent dispatch
/patient/call	LiveKit call with Ama avatar
/caregiver	Caregiver dashboard
/caregiver/summary	Session end + WhatsApp preview

Flow

Caregiver Start session with Thatha → /patient/start → Begin session → /patient/loading (POST /api/livekit/token) → /patient/call (LiveKit room) → end call → /caregiver/summary.

Deploy

Component	Platform	Command
Frontend	Netlify	npm run build
Agent	Railway	uv run agent.py start in livekit-bey-agent/

See project_overview.md for WhatsApp/webhook APIs (not yet wired). The voice stack uses Google STT + Gemini LLM/TTS + Beyond Presence (see livekit-bey-agent/).