ongrid

Ongrid is a self-hosted AIOps agent that understands your infrastructure, finds the root cause of incidents and can act, all from a chat channel. A coordinator agent dispatches to SRE, network and database specialist sub-agents; on an alert, an investigator spawns an RCA worker that walks the topology, correlates metrics, logs and traces, and writes the probable cause, down to a source-code line, back into chat.

A notable design choice is zero inbound ports: an edge agent dials out, so hosts expose no SSH, HTTP or HTTPS ports. It still offers an audited browser SSH shell through a reverse tunnel and 26+ read-only host inspection tools, and risky changes sit behind an approval and write gate. It bundles a Prometheus, Loki, Tempo and Grafana stack installed with one script, so the agent writes its own queries.

It connects to Slack, Telegram, Larksuite, DingTalk and WeCom with per-channel locale, supports Anthropic, OpenAI, GLM, DeepSeek, Gemini and Kimi with hot routing, and recently added Kubernetes lifecycle management and network device discovery. It installs from a release tarball on Ubuntu/Debian/RHEL and is AGPLv3-licensed, so network use triggers source-sharing obligations. Brand assets are separately trademarked.