This stack is the self-hosted experiment tracker and model registry boundary for Phase 92. It binds MLflow to loopback by default, persists registry metadata in PostgreSQL, and stores artifacts through MLflow's proxied S3-compatible MinIO backend. TLS, SSO, authorization, backups, and external ingress belong at the platform gateway and must be configured before production exposure.
Create the ${MLFLOW_S3_BUCKET:-mlflow} bucket before admitting runs. Supply
all required secrets through the deployment secret manager; never commit them.
docker compose --env-file /run/secrets/oshun-mlflow.env \
-f infra/nous/phase-92/docker-compose.yml config
docker compose --env-file /run/secrets/oshun-mlflow.env \
-f infra/nous/phase-92/docker-compose.yml up -d
The ExperimentBackend and ArtifactStore contracts in @nous/core are the
application boundary. Production adapters must implement authenticated calls,
bounded retries, timeouts, tenant isolation, and telemetry. The in-memory
backend is for deterministic tests and local development only.