ongrid

Ongrid is an operations-focused AI agent that aims to understand infrastructure, locate root causes, and act on them from chat channels such as Slack, Telegram, Larksuite, DingTalk, and WeCom. It is positioned as a self-hosted platform combining observability, topology, and agent-driven investigation. Key capabilities described in the README: - Coordinator plus specialist agents: a coordinator dispatches work to SRE, network, and database sub-agents. - Alert-driven auto-investigation: an investigator spawns an RCA worker and writes findings back to chat. - Root-cause analysis: walks topology, correlates metrics, logs, and traces, and links the cause to a source-code line. - Zero inbound ports: edge components dial out, avoiding exposed SSH/HTTP/HTTPS ports on hosts. - Browser SSH: reverse-tunnel shell access to hosts with auditing, without keys or a jumpbox. - One-command self-hosting via install.sh for the full stack. - Built-in observability stack wired with Prometheus, Loki, Tempo, and Grafana; the agent can write queries. - Bring-your-own model support with hot routing across Anthropic, OpenAI, GLM, DeepSeek, Gemini, and Kimi. - Two-way IM channels with per-channel locale support. - Read-only host tools: a bash sandbox plus 26+ inspection tools, with every call audited. - Kubernetes lifecycle management: enroll clusters, inspect workloads and events, manage upgrades, and mirror Kubernetes resources into topology. - Network device management: discover neighbors from edge hosts, verify with SNMP, poll interfaces, and map host-to-network-device links. Installation is documented for Linux amd64 and arm64 via release tarballs and install.sh, targeting Ubuntu 22.04+, Debian 12+, and RHEL/Rocky 9, with a CDN mirror option for mainland China. The product tour highlights root-cause analysis reports, a workflow builder, a skills catalog, MCP server registration, a knowledge vault for runbooks and repositories, an artifacts center, monitoring views, a topology map, and an approval/write gate for risky actions. Documentation is hosted at ongrid.cloud, covering getting started, install and operate, capabilities, agents, and reference material. Integrations span observability tools (Prometheus, Grafana, Loki, Tempo, OpenTelemetry, Qdrant), chat channels, and model providers. The project is licensed under AGPLv3, with brand assets under a separate trademark policy.