Introduction
ZopNight is ZopDev's Technology Value OS for AI, Cloud and Human.
ZopNight is ZopDev’s Technology Value OS for AI, Cloud and Human. Connect read-only and it prices every dollar of waste. Grant write and it fixes what you approve, inside guardrails, proved against your actual invoice.
In our experience, a cloud provider’s native advisor surfaces only a small share of the savings opportunity (roughly 14%).
Home → Overview: headline cost and savings cards above the cost trend, with detected cost anomalies called out.
What ZopNight does
ZopNight is one deterministic engine with four areas of work: optimisation, governance, AI inference, and infrastructure as code. They share one resource inventory and one audit trail, priced against your actual bill.
Optimisation engine
Five capabilities remove the waste: what to fix, when to run, how big to run, when to be ready, and rules of your own.
Recommendations
Reads your accounts, your tags and your real bill, then prices what you can stop spending: 650+ rules across AWS, GCP, Azure, Kubernetes, Databricks, Snowflake and Microsoft 365, in 11 categories, each backed by evidence.
Smart scheduling
Cron-based, per-timezone schedules run non-production resources only when your team uses them, tier-ordered or in your own order with your own delays, with production-safe overrides. 30+ schedulable resource types, including VMs, SQL, Databricks, Kubernetes and Snowflake.
Smart autoscaling
Smart Autoscaling by Tide sizes AWS Auto Scaling groups, Azure VM scale sets, GCP managed instance groups and ECS services for the moment, not for the peak: 30 to 40% compute reduction on eligible workloads. Adopt at your pace: monitor, recommend, autopilot.
Event readiness
For launches and planned spikes, ZopNight pre-scales the fleet and rolls back after, so you are ready without paying for weeks of idle headroom.
Custom recommendations
Set your own thresholds on any metric ZopNight collects. What crosses them arrives as a recommendation, like any built-in rule.
Automate → Schedules: each schedule’s next run, daily on/off window, weekly average saving, and attached resources and groups.
Governance engine
Optimisation without governance decays. Every dollar is traced to a team and a unit of value, and every change is previewed before it lands.
Showback
Spend split by team, by cloud tag and by accepted Smart Tag. Unattributed spend is tracked separately, so nothing hides.
Cost anomalies
Anomaly detection across 7 dimensions with root cause, always evaluated, with or without an alert channel.
Unit economics
Infrastructure cost per active user, per order or per 1,000 requests. Send the denominator by API, by CSV, or from an endpoint ZopNight pulls daily.
Smart Tags
Tags you never set, derived from each resource’s own metadata (provider, region, type or name), and applied back to your cloud only when you ask.
Workflow and Jira
Jira tickets from findings, with status and owner in sync both ways. Recommendations also export as an Excel workbook.
Blast radius
Every directly connected resource a fix will touch, classified At Risk (the resource is affected), Warning or Safe, with a risk score for the whole change.
AI inference
ZopNight prices the model services you already run and tells you what to switch off, and it puts a gateway in front of new LLM traffic so that traffic has a budget before it has a bill.
AWS Bedrock
Tells you which provisioned throughput, custom and imported models, and agents sit idle or would cost less on demand, and prices each against your bill.
AWS SageMaker
Tells you which endpoints take no traffic, carry more instances than they need, or never auto-stop, and which dev and test endpoints could run on a schedule.
Databricks
15 rules per cloud, so AWS, Azure and GCP workspaces are judged the same way: clusters with no auto-termination, warehouses with no auto-stop, and pools and serving endpoints holding idle capacity.
AI Gateway
One URL in front of OpenAI, Anthropic, OpenRouter and AWS Bedrock. It speaks the OpenAI SDK and the Anthropic Messages API, so most apps switch with a new base URL and key.
Keys and budgets
Each team gets an inference key with a hard dollar limit and an allowed model list, under one organisation-wide ceiling. Past 100% of its budget, a key’s calls are refused.
Routing and reporting
Everyday prompts go to the cheap model and only long or hard ones reach the strong one. Spend is reported by provider, model, tier and team.
Microsoft 365 seat licences
See which Copilot seats are used, what the unused ones cost, and which seats to reclaim.
Infrastructure as code
The same policy engine, one step earlier: ZopNight reads your own Terraform and OpenTofu on the pull request, and never runs apply.
How it runs
- Deterministic by design. Every finding ships with the metric, the threshold, and the action, and every savings finding with its dollar figure. No black-box ML. The 11 categories are idle, rightsizing, schedule, orphan, discount, compliance, security, reliability, performance, governance and advisory.
- Read-only or write, per cloud account. Connect read-only to price the waste, and grant write only for the levers you want ZopNight to pull. You can stay read-only forever, or switch an account to Read + Write when you are ready. See Read-only vs read + write.
- Databases stay with your team. Remediation never resizes, pauses, or deletes your databases (RDS, Aurora, Cloud SQL, ElastiCache, Azure SQL); database tier and replica changes stay with your team. Schedules you create can still stop and start supported databases, such as RDS and Cloud SQL, out of hours.
- Managed SaaS or self-hosted. Run ZopNight as managed SaaS, or self-host it in your own environment.
- Agentless, live in minutes. Connect read-only in about five minutes with an IAM role, service principal, workload identity, or Databricks OAuth. There are no agents to install.
- Billing-backed costs. Billed cost from Cost Explorer / Cost Management / BigQuery wherever billing is connected, and rack rate where it isn’t.
- Reports and audit. Summary, K8s, Teams, Tags, and AI Spend reports, plus the Cost Flow Sankey, four columns deep. Every change made through the API, and every MCP tool call, is captured in an audit trail you can forward to Splunk, Datadog, or any SIEM.
- Typical outcome. Customers typically take 20 to 60% off the bill in the first month.
ZopNight keeps a live inventory of 253 resource types across AWS, Azure, GCP, Databricks, Snowflake and Red Hat OpenShift, three levels deep (cluster → nodepool → VM):
| Provider | Resource types |
|---|---|
| AWS | 81 |
| Azure | 85 |
| GCP | 70 |
| Databricks | 6 |
| Snowflake | 9 |
| Red Hat OpenShift | 2 |
| Total | 253 |
Security posture
- ISO 27001:2022 and SOC 2 Type II; independently audited, attestation letter on request.
- AES-256-GCM encryption for stored credentials.
- SaaS, single-tenant, or on-prem deployment options.
- India-resident deployment in the GCP Mumbai region (MeitY-empanelled), with controls for IRDAI ICS 2023 and the DPDP Act 2023.