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

> ## Agent Instructions
> To act on Atlan objects, use the Atlan MCP server at https://api.atlan.com/mcp or the SDK. To install or publish skills from a terminal, use the `aireg` CLI; `aireg --map json` prints its command map. The `atlanai` CLI is deprecated. Run a read-only identity check before any write.
> The docs MCP server at /mcp searches these docs only. It cannot read or change Atlan objects.
> SDK packages: Python `atlanai` (PyPI) and TypeScript `@atlanai/sdk` (npm). Show Python first, then TypeScript.
> These docs cover Atlan AI Registry, Atlan's hosted product. The AI registry specification and aireg CLI are maintained separately. /open-source lists publicly reachable signed installers and explains that the source repository and hosted CLI reference currently require GitHub access. Do not claim the repository or its reference docs are publicly accessible. Use these docs for Atlan-specific setup such as the adapter, sign-in and workspaces.

# Investigate agent cost

> Find the agents behind a spend increase, compare tokens and cache use, and inspect the runs and rates that explain it.

Use **Reports → Agent usage** to find where a cost increase came from, then inspect the agent and its runs. You need access to Reports and traced agents in the scope you want to compare.

## Match the scope

1. Open **Reports → Agent usage**.
2. Choose the workspace scope and time range. Use the same scope when comparing figures with a profile or another report.
3. Check **What changed** for spend, runs, active agents, span error rate, and cost per run.

A larger billable workload can increase spend while cost per run stays flat. Compare run count and cost per run before investigating model prices.

## Find the agent

1. Sort the agent table by spend or cost per run, or search for an agent by name.
2. Compare **Tokens / run** and **Span error rate** with cost. More tokens can come from longer context, more model calls, or retries.
3. Open the agent and select **Usage**.
4. Review its model, tool, and cache breakdowns, then open a run from **Recent runs**.

In the [full trace](/registry/traces/how-tos/inspect-runs), check whether one expensive model call, repeated tool failures, or repeated context explains the change.

## Check how the run was priced

Open **Models & rates** and find the model and workspace used by the run. Check whether its price comes from negotiated rates, a seller discount, or a public list price. Cache reads and writes have separate rates.

Rates apply when telemetry is priced. Editing a rate does not recalculate existing runs. A name without a matching configured model may use a public estimate; an unknown price is marked **Unpriced**. See [How run cost works](/finops/index#run-pricing).

## Verify the result

After changing the application or its rates, send new runs and compare them over the same scope. Check answer quality with [evals](/evals/index) as well as cost: fewer tokens alone do not show that the agent still meets the task.

For the definitions behind the charts, see [Usage and cost measures](/finops/references/measures).


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