I built RuleBeat: open source, self-hosted, read-only scheduled checks for Azure governance. Feedback welcome
retained text
A while back I asked here how teams handle Azure Resource Graph queries once they stop being one-offs. The answers were genuinely useful, and most of them came down to stitching the native pieces together yourself. Disclosure up front: I was asking because I was building a tool in this space. It is public now, and this post is me showing my homework. The lesson I ended up building around: a governance query is not a governance system. A sharp query finds something real, gets shared once, and disappears into someone's terminal history. The quieter failure is the check that stops running. It does not turn red. It stays green. RuleBeat runs the governance checks your team writes for Azure on a schedule, tracks every finding over time, and never holds write access. A check is a rule you author against Azure Resource Graph or Microsoft Graph, in a visual builder or as raw KQL. Custom rules are not an add-on: they run through the same engine, scan, history, and dashboard path as the built-ins. It ships 158 checks out of the box (15 built-in rules plus the 143-rule APRL pack), findings keep their lifecycle across scans (new, active, fixed), and a suppression needs a reason and can carry an expiry date. The built-ins are the things that go quiet: unattached managed disks and unassigned public IPs you are still paying for, storage accounts allowing public blob access, app registration secrets approaching expiry, and the APRL resiliency recommendations running as part of ordinary scans instead of a once-a-year review. The deliberate trade-offs, so you know what you are getting: * **Read-only, permanently.** It never holds write credentials, never creates its own service principal, and cannot change anything in your tenant. You create the credential, so you can see in Azure RBAC exactly what it was granted. The cost is real: no one-click fix. Fixing a finding is your action, under your own identity. * **Self-hosted, one container, SQLite inside.** Nothing about your tenant leaves your deployment, and there is no telemetry. The cost: you run it. If you want to look before wiring anything up, demo mode runs the real UI over a generated database with no Azure credential at all. * **Honest numbers.** Following from the problem above: a rule that has never run, or whose last run failed, is unknown, never passing. The posture number is sometimes uglier for it, and that is the point. It is built for the point where more than one person has to know who owns a flagged row, which rows are accepted on purpose, and whether last month's fix actually held. One person with a few stable checks probably does not need it. This is v0.2, early on purpose, built by one person with AI assistance from Claude and a lot of applied AI learned along the way. There is a real list of what comes next, and what gets built first depends on what people actually hit rather than what I guess at. Feedback and suggestions are worth more to me than stars right now. Repo: [https://github.com/rulebeat/rulebeat](https://github.com/rulebeat/rulebeat) Docs: [https://docs.rulebeat.com](https://docs.rulebeat.com/) Screen-by-screen walkthrough, if you would rather look than install: [https://rulebeat.com/demo](https://rulebeat.com/demo) If you run Azure governance checks today, I would be glad if you tried it, or just opened demo mode, and told me how it went. What is missing, what you would want it to check next, what would make it worth putting in front of the rest of your team. Feature ideas are as welcome as bug reports.
document fetched 2026-08-31 00:33 · canonical https://reddit.com/r/AZURE/comments/1w2p5yy/i_built_rulebeat_open_source_selfhosted_readonly