# AI settings & usage

> Configure the AI assistant with your own provider key and scope, and track what the assistant has produced across your organization.

*[View this page in the Flax docs](https://flax-analytics.com/docs/automation-ai/ai-settings-and-usage)*

The AI assistant runs on **your own** provider key — bring your own credentials, and
Flax uses them to answer questions. This page covers configuring a key
(`/app/settings/ai`) and reviewing what the assistant has produced
(`/app/settings/ai`).

## Configuring the assistant

Add a key under **AI assistant** (`/app/settings/ai`). Each credential sets:

| Setting | Options |
| --- | --- |
| **Provider** | Anthropic (Claude), OpenAI, or xAI (Grok) — the allowed set is defined by your deployment. |
| **Model** | The provider's model id (for example `claude-sonnet-4`, `gpt-4o`, `grok-2-latest`). |
| **API key** | Encrypted at rest and never returned by the API. |
| **Scope** | *Just me* (your own key) or *Org default* (used by everyone without their own). |

At chat time Flax resolves the key in order: your personal key, then the org
default, then any server-provided default. If none exists, the assistant prompts you
to add one.

> [!IMPORTANT]
> Setting an org-default key is admin-only. Personal keys need no admin rights, and
> you can remove your own at any time.

## Tracking usage

The **AI usage** page (`/app/settings/ai`) is an admin-only view of what the
assistant has produced across the organization — output observability, not a
billing meter. It shows:

- **Graphs produced**, **distinct graphs**, and how many were **recreated** two or
  more times.
- A **Most recreated graphs** ranking (name, kind, how many times, distinct users,
  last produced) — a strong cue to save those as canonical charts so people stop
  re-asking for them.
- A **produced log** of every proposed artifact (time, who produced it, kind, and a
  spec summary), filterable by fingerprint.

> [!NOTE]
> The usage page tracks *what* the assistant proposed, not token counts or dollar
> cost. Because you bring your own provider key, billing for model usage happens
> with your provider.

## Related

- [AI assistant](/docs/automation-ai/ai-assistant) — how the assistant works.
- [Organization settings](/docs/administration/organization) — workspace configuration.