# Phase 98 AI observability

This stack runs the open-source Langfuse v3 topology: web, asynchronous worker,
PostgreSQL, ClickHouse, Redis, and S3-compatible object storage. Only the web
and media endpoints bind to loopback; databases remain isolated on the Compose
network.

1. Copy `.env.example` to `.env` and replace every placeholder. Generate
   `LANGFUSE_ENCRYPTION_KEY` with `openssl rand -hex 32`.
2. Validate with
   `docker compose --env-file .env -f docker-compose.langfuse.yml config`.
3. Start with
   `docker compose --env-file .env -f docker-compose.langfuse.yml up -d`.
4. Verify `http://localhost:3100/api/public/health` and provision SDK keys
   through the initialization variables or the Langfuse UI.

Persist and back up all five named volumes. Pin image digests during production
promotion and expose the loopback ports only through the authenticated Oshun
ingress. Langfuse traces can contain prompts, retrieved context, and user data;
apply the same retention, access-control, and redaction policy as production AI
decision logs.
