zopnight

We shipped custom recommendation policies.

You set it. Pick the resources you care about, pick the metric to watch, set your threshold and the time window, and tell Zopnight what to suggest when it gets crossed. A few clicks, no YAML, no support ticket.

You pick the action, not just the alert. Stop something that's idle, schedule it off during quiet hours, or remove it if you're done with it. The recommendation shows up with the fix already attached.

Spikes count, not just averages. Averages hide things. A rule like "flag anything whose CPU ever crossed 80%" works now, so a short burst at 3am doesn't get smoothed away into looking healthy.

Still gated. Every custom recommendation explains why it fired, shows the savings you can expect, and warns you before you switch off anything critical like a database. Nothing moves until a human says yes.

Your rules sit right beside our built-in ones and behave identically. Same explanation, same one-click fix.

Why we bothered: our thresholds were ours. One team's idle is another team's batch job, and we kept tuning a number that was never going to be right for everybody. Handing over the knob was cheaper than guessing, and the rules people write themselves are the ones they actually act on.

You can poke at it in the Playground, no login needed: https://zop.dev/zopnight/playground

Curious where you land on this one. Do you ship opinionated defaults and defend them, or open up config early and let users build their own rules?

I used to think the hard part of this was finding the wasted cloud spend. It isn't. A dashboard can do that. We had it working early and felt pretty good about ourselves.

Then a customer asked the question that stalled us for a month: "Okay, you found $40k of idle stuff. Now what, I go delete 200 resources by hand?"

That's the real problem. Finding waste is just a dashboard. Fixing it is a 2am incident waiting to happen. One wrong delete to save $12 a month and you've taken down production.

So we stopped building "find more waste" and spent a few weeks on the part nobody demos: making it safe to actually act. Here is what shipped in Zopnight.

1. Cost Anomaly Detection. Instead of a scary red number, Zopnight tells you why the bill moved. "Reservation expired, switched to On-Demand." "Schedule WorkHours may have failed." Nobody acts on an alert they don't understand.

2. Blast Radius. Before anything gets touched, ZopNight shows you what breaks. Delete a schedule and it doesn't just say "done". It shows the 15 servers that will now run 24/7 and your savings dropping from 42% to 0%, in real dollars. You see the damage before you cause it.

3. Auto-remediation. Zopnight runs 450+ checks across AWS, GCP and Azure, but every automatic fix stops at a human approval step. Nothing moves until someone says yes. We spent longer building that gate than we did building the fixes.

4. Governed MCP. You can now let an AI agent in Cursor or Claude Code start, stop and deploy through Zopnight. The rule is simple. An agent can never do more than the person whose token it is using, and writes are off by default. We wrote the risk into our own docs instead of hiding it.

5. Showback, the payoff. This is the part that finance guy actually asked for. Every dollar now rolls up to the team that spent it. Zopnight ties cost to real business numbers like cost per 1,000 orders or monthly active users, and it can even tell you who created each resource, pulled from your cloud activity logs. So every dollar has an owner.

Honestly, the hard part was never the finding. It was earning enough trust that a real team would let software touch their cloud at all.

If you want to poke at it, you can try Zopnight live in our Playground, no login needed: https://zop.dev/zopnight/playground

So where do you land? Would you let a tool or an AI agent auto-fix your infra, or does it always need a human on the button?

• Muskan_z

Wanted to share what we've been building, and the one question that started it.

Last quarter, someone in finance asked us: of everything we spend on cloud and AI, which parts are actually creating value? We had dashboards for how much we spent. Not one of them could tell us whether it was worth it. That gap bugged us enough to build around it.

When teams run real infrastructure across AWS, GCP, and Azure, spending gets hard to control. Resources sit idle nights and weekends, ownership is unclear, and the path from "this looks wasteful" to "it's actually fixed" runs through spreadsheets and risky manual changes.

What is ZopNight?

ZopNight is a Technology Value OS for AI, Cloud, and Humans. You connect AWS, GCP, and Azure read-only, and it runs one continuous loop: find, own, fix, prove:

• Finds the waste your dashboards miss (idle resources, over-provisioned instances, orphaned storage, the cross-cloud stuff nobody checks).

• Route ownership so every resource has a name next to it, and nothing stays "nobody's problem."

• Automates the boring optimisation (schedule non-prod off, right-size, prep for spikes), with one-click fixes behind an approval gate.

• Proves which spend is actually creating value, including your AI and LLM bill, tied back to teams and outcomes.

Connect your accounts, run the audit, and see the number. Nothing moves until you say so, and ZopNight never touches your databases.

Why it exists

Every cloud already ships its own advisor (Trusted Advisor, GCP Recommender, Azure Advisor), but each one only sees its own cloud and a fixed checklist. None of them tells you who owns a resource or whether it's worth keeping. We wanted one place that spans all three, plus the AI runtime, and that we'd actually trust near production.

Who is it for?

Engineering, platform, and FinOps teams running production infrastructure on AWS, GCP, or Azure who want to cut idle spend, attribute every dollar, and optimise safely without slowing down the people shipping.

If you've ever been the person who had to explain the cloud bill, you'll recognise a lot of this. Curious how others here are tracking cloud + AI value today?