> ## Documentation index
> Fetch the complete documentation index at: https://rolter-ai-rolter.docs7.io/llms.txt
> Ask it with GET https://rolter-ai-rolter.docs7.io/ask?q=<keywords or question>. Short keywords rank best.
> Use these to discover all available pages before exploring further.

# Tracking and Capping LLM Spend

> Make every request's cost trustworthy, see spend by model, team, business unit and customer, stop it at limits you choose, and give finance read access without exposing prompts.

Rolter records the cost of every request it serves, attributes it to the key, project, team, business unit and customer behind it, and can stop traffic at a budget. This guide sets that up in the order that makes the numbers trustworthy.

<Note>
  Budgets and rate limits keep their counters in Redis, so the limits hold across every gateway replica. Without Redis they fail open: requests pass and spend is not counted. Cost reporting itself comes from the request logs in ClickHouse.
</Note>

<Steps>
  <Step title="Price every model that costs money">
    A model with no price row is logged at a cost of **zero**, which is indistinguishable from free. Add prices under **Models → Model Prices** (or `[[model_prices]]` in the config file), in any currency your rate table knows. See [Budgets & rate limits](/security/budgets-and-rate-limits).
  </Step>
  <Step title="Find what is still unpriced">
    **Observability → Dashboard** shows how much of the window's traffic was unpriced, and **LLM Logs** flags each such row, so a partial total is never read as a final one. Price what shows up, or decide per budget whether unpriced traffic counts as zero or is refused.
  </Step>
  <Step title="Attribute spend to the people you bill">
    Create **business units** and name the **customers** you serve under **Governance → Business Units** and **Governance → Customers**, then point each virtual key at a unit and a customer. Every request the key serves is recorded against both, and the same screens roll the spend up. See [Cost attribution](/security/cost-attribution).
  </Step>
  <Step title="Set the limits">
    **Models → Budgets & Limits** at the org, team, project, key, business unit or customer:

    - a **budget** caps spend over a day, a month or for good; the next request past it gets HTTP 402, and nothing reaches the upstream;
    - a **rate limit** caps requests and tokens per minute; past it, HTTP 429 with `Retry-After`.

    The most restrictive limit on a request's chain wins. To change a budget, replace it: spend so far is counted per scope, so the new budget starts from what was already spent.

  </Step>
  <Step title="Get alerted on runaway spend">
    **Alerting → Alert Rules** with the `spend_velocity` metric sends a webhook when spend per hour crosses the line you set, alongside error-rate and latency rules.
  </Step>
</Steps>

## Give finance access without the prompts

Make your finance team **viewers** of the org. A viewer sees the Dashboard, every request-log row of the org — model, tokens, cost, status, attribution — and every budget and limit, and can change none of them. The captured request and response bodies are shown only to the members and admins of each project, unless a project admin chooses to open them to its viewers. See [RBAC](/security/rbac#who-sees-which-request-logs).