> ## Documentation Index
> Fetch the complete documentation index at: https://docs.costoptix.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Recommendations

> Savings opportunities across cloud, Kubernetes, hosts and AI plans, with savings checked against the bill after you act.

## Overview

Open **Optimisation** in the navigation. CostOptix analyses your spend every night at 05:00 UTC and lists what you can act on, with the projected monthly saving for each finding.

Every finding comes from a fixed set of rules applied to your billing and usage data. The AI assistant can explain findings, but it never invents them.

***

## Tabs

| Tab | What it finds | Scope |
| - | - | - |
| **Savings** | Every saving for the selected account, from all engines | Selected account |
| **Orphaned** | Resources attached to nothing that still bill | Every Azure account |
| **Commitments** | Services billing steadily on demand with no commitment applied | Selected AWS or Azure account |
| **Kubernetes** | Pod rightsizing, node sizing and idle capacity | Selected cluster |
| **Hosts** | Idle services on a machine | Selected host |
| **AI plans** | Idle seats and plans that do not fit how people use them | Your LLM plans |
| **Anomalies** | High and critical anomalies from the last 90 days, to investigate | Selected account |

Filter by engine, status, provider or text, and export what you see as CSV.

### Orphaned resources

Azure only. CostOptix uses Azure Resource Graph to find resources that are attached to nothing but still cost money, including:

* Unattached disks and stale snapshots
* Unassociated public IPs
* Empty load balancers, application gateways and Traffic Manager profiles
* Idle App Service plans, virtual network gateways and empty SQL elastic pools
* Unused NAT gateways, DDoS plans and Front Door WAF policies
* Disconnected private endpoints and unlinked private DNS zones

Savings here are observed, not modelled. Open **Group by type and clean up in bulk** to work through them by resource type.

### Commitments

AWS and Azure. CostOptix looks for spend that is steady enough to commit to and is still billed at on-demand rates. It works in commitment pools rather than per service, because one savings plan covers every eligible service in its scope.

* Where the provider gives its own purchase recommendation (Azure savings plans), that figure is used.
* Otherwise CostOptix uses a conservative floor: a low percentile of recent daily spend, priced at a published minimum discount.

<Note>
  CostOptix recommends commitments but never buys them. Purchases stay in your cloud console.
</Note>

### Kubernetes

From the sealed daily ledger, not live metrics:

* **Pod rightsizing.** Workloads requesting more CPU or memory than they use.
* **Node sizing.** Whether fewer or different nodes would carry the same load.
* **Idle capacity.** Node capacity no workload has claimed.

### Hosts

Services that sit idle on a machine, measured from the host agent.

### AI plans

Idle seats and people who would be better off on a higher tier, a lower one, or pay-per-token. Based on coding-agent telemetry only, so chat apps and tools that send no telemetry are not counted. Check with the person before removing a seat. USD plans only.

***

## Working a finding

Each finding has a status:

| Status | Meaning |
| - | - |
| **Open** | New or still applies |
| **Snoozed** | Hidden for now, comes back later |
| **Done** | You made the change |
| **Dismissed** | Not relevant |

Open a finding to see when it was first detected, when it was last confirmed, and the evidence behind it.

***

## Verified savings

Marking a finding **Done** does not count the saving straight away. CostOptix waits, then compares billed cost before and after the change, using the same sealed data the monthly bills are built on. A projected number never counts as saved.

| Badge | Meaning |
| - | - |
| **Verifying** | Waiting for enough billed days after the change |
| **Verified** | The saving showed up on the bill |
| **Unconfirmed** | The data could not confirm a saving either way |
| **Cost rose** | Cost went up after the change |

How it is checked:

* **Kubernetes rightsizing, orphaned resources and commitments** compare 14 billed days before the change with 14 days after. Cloud findings also wait out the provider's billing lag so restated figures are not measured.
* **AI plans** are checked two weeks after you mark them done, against your plan settings.
* Savings under \$5 a month are not verified, to keep noise out.
* A finding stops being re-checked 90 days after it was marked done.

The summary at the top of each tab shows open projected savings and verified savings. When nothing is verified yet, it shows implemented savings and says they are projected.

***

## Related

<CardGroup cols={2}>
  <Card title="Kubernetes" icon="dharmachakra" href="/providers/kubernetes">
    Waste & fixes, ranked for you.
  </Card>

  <Card title="Hosts" icon="server" href="/providers/hosts">
    Right size and what-if pricing.
  </Card>

  <Card title="Anomaly Detection" icon="bell" href="/features/anomaly-detection">
    Where the anomaly findings come from.
  </Card>

  <Card title="AI coding tools" icon="robot" href="/providers/llm">
    Plans, seats and usage value.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.