Nightingale open-source alerting platform ships built-in AI agent and MCP server

Nightingale is an Apache-2.0 self-hosted alerting platform that centralizes alert rules, incident handling, and notification routing, querying existing Prometheus, VictoriaMetrics, Elasticsearch, Loki, and ClickHouse data sources without replacing your collectors, storage, or Grafana dashboards. Rules can be split by business group with team permissions, login supports OIDC, OAuth2, LDAP, and CAS, and notifications cover Slack, PagerDuty, Discord, email, Telegram, Mattermost, Jira, plus HTTP webhooks and scripts. The highlight is its built-in AI agent, usable directly in the Web UI: it can call tools to inspect monitoring data and configuration, query related metrics and logs in multiple steps, diagnose why alerts did not fire or were not delivered, generate PromQL and SQL, and create or update alert rules and dashboards. Changes to existing rules and dashboards are proposed first and wait for confirmation. The agent supports OpenAI-compatible APIs, Claude, Gemini, and compatible self-hosted models, with users configuring model services and credentials themselves; core alerting runs independently of the agent. The platform also offers an MCP server (path /mcp) that exposes read-only tools by default under the caller's existing API permissions, with write tools requiring explicit configuration; A2A lets external agents delegate tasks to the built-in agent. For ops and SRE teams running self-hosted monitoring stacks, this means converging alerting onto one platform without migrating data and delegating troubleshooting and configuration to an agent backed by self-hostable models; production deployments should still pin versions, configure credentials and TLS, back up the metadata database, and monitor rule evaluation and notification failures.