Spring Boot · Distributed Systems · Safe AI

Sentinel responds to incidents without surrendering control.

Alerts are deduplicated, durably queued, correlated with operational evidence, and evaluated by a grounded agent. Every proposed mutation must still pass deterministic Java guardrails, idempotency checks, and an append-only audit ledger.

Checking application readiness…
01 / RELIABILITY

At-least-once, safely

RabbitMQ manual acknowledgements combine with PostgreSQL uniqueness and idempotent consumers so redelivery cannot duplicate an incident.

02 / GROUNDING

Evidence before proposals

Bounded deploy, metric, log, and pgvector runbook retrieval ground model output. Missing evidence escalates instead of inventing a fix.

03 / SAFETY

The model cannot authorize

A single deterministic gate enforces kill switch, allowlists, risk, dry-run, approval, and action idempotency before any strategy can execute.