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

# AI Recommendations

> Measured model, seat, and cap findings across Bedrock, Cursor, and Claude Enterprise, plus Clara's scheduled analysis of your AI spend.

**Recommendations** is the AI slice of nOps recommendations: concrete, measured findings about model choice, unused seats, and oversized spend caps, grouped by the rule that produced them. Below the findings sits **AI Analysis**, where Clara's scheduled prompts about AI Services report back.

<Note>
  Recommendations is available to anyone with AI access. No seat tracking project is needed.
</Note>

## Get started

1. Sign in to [nOps](https://clara.nops.io/dashboard).
2. In the sidebar, open **AI → Recommendations**.
3. Start with the largest group, expand it, and work down.

## Key features

### Summary cards

| Card                          | What it counts                               |
| ----------------------------- | -------------------------------------------- |
| **AI recommendations**        | Open findings on your AI spend.              |
| **Estimated monthly savings** | Cash off the invoice. Excludes cap headroom. |
| **Cap headroom**              | Unused spend-limit exposure, not cash.       |

The split between the last two cards matters. Lowering an unused spend cap reduces risk but changes nothing on the bill — the provider charges actual usage — so headroom is reported separately rather than inflating a savings number you will never see materialize.

### AI Optimization Recommendations

The main list groups findings by rule, with a count and a total per group. Expand a group to see individual rows. Each row shows the recommended action, the owner, the owner type, the tool it applies to, and either an estimated monthly saving or a **headroom** figure, plus an effort badge.

Filter by tool with the selector in the section header — **All tools**, **Bedrock**, **Cursor**, or **Claude Enterprise**. Each option carries its own count, so you can see where the findings are before switching.

### AI Analysis

The lower section surfaces Clara scheduled prompt results scoped to **AI Services**. This is narrative analysis rather than rule output: whatever your scheduled prompts about AI spend turned up, most recently first.

## FAQs

<AccordionGroup>
  <Accordion title="Why does a finding show 'headroom' instead of a dollar saving?">
    Because it is a spend cap, not usage. The cap is a guardrail; lowering an unused one removes exposure without reducing the invoice. Those figures are deliberately excluded from the savings total.
  </Accordion>

  <Accordion title="How do these differ from the main Recommendations page?">
    They are the same data. This page pre-filters to AI findings and adds the AI-specific summary cards and tool filter, so you do not have to dig them out of a list dominated by storage and compute.
  </Accordion>

  <Accordion title="Nothing shows for a tool I use.">
    A tool only produces findings once nOps has enough measured usage for it. Check that the integration is connected and syncing.
  </Accordion>

  <Accordion title="How do I get more AI Analysis entries?">
    Schedule Clara prompts against the AI Services domain. Their results land in this section.
  </Accordion>
</AccordionGroup>

## Related

* [AI Anomalies](/ai/anomalies)
* [Tracking](/ai/tracking)
* [Insights](/inform/insights)
* [Amazon Bedrock integration](/integrations/amazon-bedrock)