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

# Reporting

> See what your AI traffic costs and where the month is heading, how often your guardrails fire, and which MCP tools are used.

**Reporting** answers three questions about your traffic through the [AI Gateway](/ai-gateway/overview.md): what does it cost, what are the guardrails catching, and which tools are being used. Each has its own tab. Owners, admins and Governance Leads can open it.

The cost report also appears on each AI system's and each [project's](/projects.md) **Cost and usage** tab, for just that AI system or project.

## Cost and usage

The figures at the top show spend, requests, tokens and cache savings for the range you pick, each compared with the same length of time before it.

**Projected this month** shows where this month's spend is heading at its current pace. When a monthly budget applies, the projection is drawn against it and says **Over** if it will pass it. Select it to open [Budgets](/ai-gateway/budgets-and-limits.md#see-every-budget). The projection starts on the 3rd of the month, once there is enough to go on.

**Cache savings** estimates what the requests answered by the [semantic cache](/ai-gateway/semantic-cache.md) would have cost.

Point at a figure with a dotted label to see how it is worked out.

Below the figures, the **Spend** chart shows spend over time, and the rankings show who and what spent it: by project, AI system, person, model, provider, client or key. The ranking you pick also splits the chart. **No AI system** collects traffic on organization and personal keys.

## Guardrail violations

How often your guardrails fire, where, and what they do about it. Only counts are shown, never the text that was caught.

The main figure is **Violations per 1,000 checked**: of every thousand requests the guardrails checked, how many something fired on. Requests no guardrail checked are left out, so a busy, unguarded AI system does not make the rate look better. Each request with a violation counts once, as **Blocked**, **Masked** or **Logged only**.

The rankings put the highest rate first, so a small AI system that trips often is not hidden under a busy one. **What was found** names the kinds of thing caught, such as **Email address**.

Two sections help you decide what to tighten:

* **Logged only** lists guardrails that fired but only logged, so you can see whether to switch them to **Block**.
* **On tool calls** lists guardrails that fired on [MCP tool](/ai-gateway/mcp.md) arguments or results.

Select any row to open the matching requests in the [request log](/ai-gateway/request-logs.md). On an AI system's row, **Create incident** opens an [incident](/incidents.md) linked to this report.

## Tool use

Which MCP tools are called, by which AI systems and people, and how often they fail. Tools the server marks [destructive](/mcp-servers.md#what-each-tool-can-do) are listed first. A call a guardrail blocked does not count as a failure. Select a row to see its calls in the request log.

## Pick a range and filter

Choose **Today**, **7 days**, **30 days** (the default), **This month**, **Last month** or **This quarter**, or pick your own days, up to 92. Days run midnight to midnight in your browser's time zone. Once your organization has projects, **Project** narrows every tab to one.

The tab, range and filters are in the page's address, so a link you share opens the same view.

## Export and print

**Export CSV** saves the chart or the ranking on screen as a spreadsheet file, named after what it holds and the dates. **Print** prints the report alone, in black on white, with its range and filters. Both happen in your browser: nothing extra is sent to Altrum.


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