The stack runs without a key (collection, queries, alerting); the agent needs one.
git clone https://github.com/Joshwong1908/relionaut.git && cd relionaut cat >> .env <<'EOF' GLM_API_KEY=your-key # any Anthropic-compatible endpoint GLM_BASE_URL=https://open.bigmodel.cn/api/anthropic GLM_MODEL=glm-4.6 EOF
docker compose up -d --build
Open http://localhost:8082. Within 30 seconds the dashboard shows live telemetry; the alert engine evaluates rules every 60s, opens incidents, and — with a key configured — runs an automatic investigation on each new incident.
Open any incident and chat: the agent investigates with read-only tools (query_logs, query_metrics, whitelisted host commands, HTTP probes) and streams its reasoning. Conclusions arrive through submit_findings with verbatim evidence quotes. After a run, distill it into a lesson with POST /api/incidents/:id/distill.
| Service | Port | Role |
|---|---|---|
| nginx | 8082 | static console + /api proxy |
| api | internal | Hono routes, alert engine, investigation agent |
| postgres | internal | incidents, trajectories, telemetry |
| redis | internal | rate limiting |