An open-source PagerDuty+Rundeck clone that keeps LLMs on a leash

What it does

ITOps Agent Platform is a web-based IT operations center built in TypeScript and React that wires multiple LLM-powered agents into visual workflows. It ingests alerts from sources like Zabbix and Prometheus, runs automated diagnosis and log analysis, generates remediation scripts, and pushes them to a mobile approval gate before execution. The goal is a closed loop from alert to verified fix while keeping a human in the chain.

The interesting bit

Most “AIOps” projects either dream of full autonomy or just chat with logs; this one treats human approval as a first-class constraint. It pairs a multi-agent reasoning pipeline with command safety filtering and audit trails, acknowledging that letting an LLM restart a database unsupervised is still a bad idea.

Key highlights

• Claims 12 specialized agents, 68 API routes, 72 services and 10 visual workflows (per README badges)
• Supports both international and domestic Chinese LLM backends
• Explicit human-in-the-loop approval with mobile push and full execution audit
• Targets Zabbix/Prometheus alert ingestion with automated root-cause analysis
• Self-hosted Docker deployment intended, though currently discouraged by the maintainers

Caveats

• The codebase is undergoing a major refactoring to 4A/DDD architecture; the README explicitly warns that recent Docker images are transitional, dependency-chaotic, and “not recommended to pull directly”
• Ordinary users are told to wait for a stable release, while only developers comfortable debugging source builds are invited to try it now
• The project has 543 stars and a very loud README; the “$400 billion market” claims are aspiration, not traction

Verdict

Worth watching for SRE teams at small-to-mid-size shops that want an open-source alternative to PagerDuty and Rundeck stacks, especially if they run Zabbix or Prometheus. Skip it if you need something stable today; the maintainers themselves say it is not ready for ordinary users yet.