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

# Control How Your Team Uses AI (AI Policy)

> Set your Elephas Team AI policy: turn cloud AI off, allow it, or allow it only with redaction. Set usage caps per member. Personal API keys never apply here.

<Warning>
  Draft. This article depends on an open product question and is not ready to publish.
</Warning>

The Team Owner decides how everyone in the Team workspace uses AI. You can turn cloud AI off, allow it, or allow it only after personal details are replaced with placeholders. You can also cap how much of the Team's AI allowance each member uses.

<Info>
  Available in **Team workspaces**. Only the **Owner** can change the policy. Owners and Admins can view usage and set caps.
</Info>

## Before you start

* You are the Team Owner. Admins and Members can open **AI Policy** but cannot save changes: "Only the Team Owner can change the AI policy."
* An internet connection. The policy is stored on the Elephas server and applies to every member's computer.

## Steps

<Steps>
  <Step title="Open AI Policy">
    Go to **Team Settings › AI Policy**.
  </Step>

  <Step title="Choose a Cloud AI mode">
    Under **Cloud AI mode**, pick **Disabled**, **Allowed** or **Allowed with required redaction**. The modes are explained below.
  </Step>

  <Step title="Set a default cap (optional)">
    Under **Limits**, enter a number in **Default cap per member (tokens per period)**. Leave it empty for no cap.
  </Step>

  <Step title="Save">
    Click **Save Policy**. Elephas confirms "Policy saved." and shows the new policy version.
  </Step>
</Steps>

## How it works

### The three modes

| Mode | What members get |
| - | - |
| **Disabled** | "Cloud AI is off for everyone in this Team." Local AI is off in this Team workspace too. |
| **Allowed** | "Cloud AI is allowed for every member." Questions go to Elephas AI through the Elephas gateway. Local AI can also be used. |
| **Allowed with required redaction** | "Cloud AI is allowed once context has been redacted." Elephas replaces personal details with placeholders on the member's computer before anything is sent. If that can't happen, nothing is sent. Local AI can also be used. |

When a member asks with cloud AI turned off, they see: "Your Team's policy has cloud AI turned off, so this request was not sent."

### Personal information protection in a Team

In a Team workspace, the Owner's policy decides whether personal details are replaced before cloud AI, not each member's own setting. Under **Allowed**, questions and passages are sent without that step. Under **Allowed with required redaction**, it always runs. See [Personal information protection](/docs/privacy-and-security/personal-information-protection) for the kinds of details it hides, and [What data is sent to AI](/docs/privacy-and-security/what-data-is-sent-to-ai) for what leaves the computer.

There is also a box, **Run the gateway residual check over redacted context**. It asks the Elephas gateway to run an extra check on the already-redacted text.

### Where Team AI runs

Team cloud AI always goes through the Elephas gateway. Personal API keys never apply in a Team workspace, even for the Owner, and even if you saved a key in your Personal workspace. The gateway does not store prompts or answers. In the app's words: "Retention and training use by the upstream and its vendors follow their own policies." The AI Policy screen names the model provider.

With Local AI, a model on the member's own computer answers, and the question and passages are not sent to any AI provider. See [Run Local AI](/docs/ai-options/run-local-ai).

### Usage and caps

Team cloud AI draws from one pooled allowance for the whole Team, sized by the number of paid seats. Owners and Admins see who used how much in the **Elephas AI usage** panel on the AI Policy screen.

* The **default cap** applies to every member.
* To give one person a different limit, enter it in **Cap (tokens)** on their row and click **Set Cap**. **Remove Cap** clears it.
* A member who reaches their cap sees: "You've reached the limit your Team set for you." Caps reset with each usage period.

## Troubleshooting

<AccordionGroup>
  <Accordion title="The policy changed on the server since you loaded it">
    Someone saved a change at the same time. The screen reloads the current policy. Review it and save again.
  </Accordion>

  <Accordion title="Members see a refusal about redaction">
    "Your Team's policy requires redaction before cloud AI, and no redaction engine is available on this device, so nothing was sent." The member's computer may still be downloading what it needs. See [Personal information protection](/docs/privacy-and-security/personal-information-protection). Or switch the mode to **Allowed**, or ask members to use Local AI.
  </Accordion>

  <Accordion title="Your Team's AI policy just changed">
    The policy changed while a member was asking. They can ask again and the new policy applies.
  </Accordion>
</AccordionGroup>

More fixes in [Team errors](/docs/troubleshooting/team-troubleshooting).

## FAQ

<AccordionGroup>
  <Accordion title="Can a member use their own OpenAI or Claude key in the Team?">
    No. Team workspaces never use personal API keys. A member's key keeps working in their own Personal workspace.
  </Accordion>

  <Accordion title="Can I turn cloud AI off for one client's Brain only?">
    No. The policy covers the whole Team workspace. For one sensitive matter, members can switch to Local AI.
  </Accordion>

  <Accordion title="Does the Team usage come out of my Personal plan?">
    No. Team questions use the Team's pooled allowance only, and Personal questions use your Personal plan.
  </Accordion>
</AccordionGroup>

## Related articles

<CardGroup cols={2}>
  <Card title="What data is sent to AI" href="/docs/privacy-and-security/what-data-is-sent-to-ai" />

  <Card title="Personal information protection" href="/docs/privacy-and-security/personal-information-protection" />

  <Card title="Seats and billing" href="/docs/teams/team-seats-and-billing" />

  <Card title="Confidential client work" href="/docs/privacy-and-security/confidential-client-work" />
</CardGroup>


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