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

# AI usage and credits

> Use DataFramer credits or your own keys

Prices and plan limits are on the [pricing page](https://dataframer.ai/pricing).

## How a model call is paid for

Judges, trace analysis in Findings, Review Copilot and generation all call models. Each call is paid in one of two ways:

* **Your own model credential**, when a company admin has saved one under **Settings → API Keys**.
* **DataFramer credits**, when no credential is saved. A company admin buys them under **Settings → Billing**.

With neither, DataFramer refuses new AI tasks and asks you to add a credential or buy credits.

## Your own model credential

A company holds one credential at a time: an Anthropic API key, an OpenAI API key, or, on Enterprise, a Databricks service principal. Saving one for a different provider replaces it.

While it is saved, every model call runs on that provider. When a task requires a model from another provider, DataFramer runs the equivalent model of your provider instead. When there is no equivalent, the task is refused: choose another model, or replace the credential.

## The monthly spend limit

Your plan caps your company's AI spend per calendar month.

* It counts every model call, including calls on your own credential.
* At the limit, DataFramer refuses new AI tasks until next month. Your data stays readable, and an **Auto refresh** on [Tracking](/docs/findings/discovery-and-tracking) skips its runs and resumes by itself next month.
* For a higher limit, move to a higher plan.

Company admins see this month's spend against the limit under **Settings → Billing**.

## Work that fails

Failed work does not count toward the spend limit and uses no credits. On your own credential, your provider still bills those calls.


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