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

# The AI engine

> The model that powers chat and insight explanations — and the evidence guardrail

Koala uses a language model to **explain** your cost data — to narrate insights and answer chat questions in plain language. The model is configurable per tenant.

## Two ways to power the AI

You configure this in **Settings → AI Providers**, and there are two modes:

<CardGroup cols={2}>
  <Card title="SaaS mode" icon="cloud">
    Use the **Koala-provided** model. Nothing to set up — usage is **capped by your plan**, and a banner shows tokens used, your limit, and tier. When you reach the cap, Koala prompts you to add your own key.
  </Card>

  <Card title="Bring your own key" icon="key">
    Add your **own provider key** (OpenAI by default) for **uncapped** usage on your own account. Manage, test, and reset usage from the same tab.
  </Card>
</CardGroup>

| Setting      | Default                                                           |
| ------------ | ----------------------------------------------------------------- |
| **Provider** | `openai` — swappable behind a common interface.                   |
| **Model**    | `gpt-4o-mini`.                                                    |
| **Base URL** | `https://api.openai.com/v1` (override for a compatible endpoint). |

<Info>
  Your API key is **encrypted at rest** and shown only as a masked hint. Use **Test** to validate it, and view or export **token usage** from the AI Providers tab.
</Info>

## The evidence guardrail

The most important thing to understand about Koala's AI is what it does **not** do:

<Warning>
  **The model never invents numbers.** Every figure in an insight or a chat answer comes from SQL over your own data. The language model only turns those verified results into readable explanations — it is given the evidence and asked to explain it, never to produce the numbers itself.
</Warning>

This is why Koala's answers are safe to act on: each one can be traced back to the exact query and aggregates that produced it. See [How it works](/finops/how-it-works) for the full pipeline.

## Next steps

<CardGroup cols={2}>
  <Card title="Ask Koala AI" icon="comments" href="/finops/chat">
    Query your cloud costs in plain language.
  </Card>

  <Card title="Insights" icon="lightbulb" href="/finops/insights">
    Evidence-backed findings and recommendations.
  </Card>
</CardGroup>