> ## 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.

# What Data Elephas Sends to AI (Elephas AI, Own Key and Local)

> Exactly what Elephas sends when you ask a question with Elephas AI, your own API key or Local AI, where it goes, and what stays on your computer in every mode.

Elephas never uploads your files or your library to an AI. When you ask a question, it sends only your question, a few recent chat messages, and the passages from your documents needed to answer. Where that goes depends on the AI you choose: the Elephas gateway, your own provider, or nowhere at all with Local AI.

## You choose where AI runs

In **Settings › AI**, under **How Elephas answers**, you pick one of three options:

* **Elephas AI (included with your plan)**
* **My own AI account**, which uses your own API key
* **Local (on this computer)**, which is Local AI

Every answer shows which one answered, for example **Answered by: Elephas AI**. See [Choose how Elephas answers](/docs/ai-options/choose-how-elephas-answers).

## What is sent, by mode

| | Elephas AI | Your own API key | Local AI |
| - | - | - | - |
| **Where it goes** | The Elephas gateway, which forwards it to a third-party model provider | Straight from your computer to the provider you chose | Nowhere. A model on your computer answers. |
| **Your question** | Sent | Sent | Stays on your computer |
| **Recent chat messages** | Up to the last 4 messages, size-limited | Up to the last 4 messages, size-limited | Stays on your computer |
| **Passages from your documents** | The passages Elephas finds for the question, which can include passages the final answer does not use | Same as Elephas AI | Stays on your computer |
| **Document names** | File names (no folder paths) or web page titles of documents in scope, and section headings | Same as Elephas AI | Stays on your computer |
| **Brain names** | Not sent. Only the number of Brains and documents. | Not sent | Stays on your computer |
| **Whole files or your library** | Never | Never | Never |
| **Personal information protection** | On by default | On by default | Not needed, nothing is sent |
| **Who can identify you** | Elephas, through your sign-in. The model provider sees requests from Elephas. | Your provider, through your API key | No one |

The search index, your full documents and your other chats are never sent in any mode. **Indexing happens on your computer. Always.**

## How it works

### How a question is answered with cloud AI

With Elephas AI or your own API key, Elephas works in steps. It searches your documents on your computer, sends the best passages to the AI, and the AI can ask Elephas to look up more: search again, list the documents in scope, or read a section. Each lookup runs on your computer, and only the results are sent. So one question can mean several requests, each carrying passages from your documents.

Before each request, if **Protect personal information when using cloud AI** is on, Elephas replaces many common personal details with placeholders and restores them in the answer on your computer. See [Personal information protection](/docs/privacy-and-security/personal-information-protection). The shield under the answer lists what was hidden. See [Review hidden details](/docs/privacy-and-security/review-hidden-details).

### Elephas AI

When you choose **Elephas AI**, Elephas tells you: "Your question and the selected sources leave this computer for the Elephas gateway, which forwards them to" a named third-party model provider. "Cloud inference is not end-to-end encrypted."

* **The gateway does not store your prompts or answers.** It logs only IDs, counts and decisions, such as how many tokens a request used, so your plan's AI allowance can be measured.
* **The model provider's terms apply.** In the app's words: "Retention and training use by the upstream and its vendors follow their own policies." The provider's name is shown on the Elephas AI card in **Settings › AI**.
* No personal API key is used on this route.

See [Elephas AI](/docs/ai-options/elephas-ai).

### Your own API key

When you choose **My own AI account**, Elephas tells you: "With your own key, your question and the selected sources go straight to" your provider. "Elephas is not involved and cannot see them."

* Requests go directly from your computer to the provider you picked: OpenAI, Anthropic (Claude), Google Gemini, Groq, Mistral, xAI (Grok), DeepSeek, OpenRouter, Together AI, or a custom endpoint you set up.
* Your API key is stored in Keychain (Mac) or Credential Manager (Windows), and never synced.
* Your provider's API terms decide whether your data is retained or used for training. Check them for the provider you use.
* Elephas still checks your plan online before it uses your key. That check sends only a device ID, never your key, question or documents.

See [Use your own API key](/docs/ai-options/use-your-own-api-key).

### Local AI

With **Local (on this computer)**, answers are generated by a model running on your computer. In Local AI mode, your question, your documents and your chats are not sent to any AI provider, and no AI request leaves your computer.

Elephas still contacts the Elephas service to confirm your plan, at most once an hour while you use it. That check sends only a device ID. See [Use Elephas offline](/docs/privacy-and-security/use-elephas-offline) for everything that still goes online.

### Team workspaces

In a Team workspace, the Team Owner's AI policy decides what is allowed: cloud AI disabled, allowed, or allowed only with redaction. Team cloud AI goes through the Elephas gateway, and personal API keys do not apply. See [Team AI policy](/docs/teams/team-ai-policy).

## Troubleshooting

<AccordionGroup>
  <Accordion title="Elephas can't protect personal information until it finishes downloading">
    Protection is on, but its models are not ready, so nothing was sent. Finish the download in **Settings › Downloads**. See [Personal information protection](/docs/privacy-and-security/personal-information-protection).
  </Accordion>

  <Accordion title="I'm not sure which AI answered">
    Look at the badge on the answer: **Answered by:** followed by the mode or provider. Change it in **Settings › AI**.
  </Accordion>
</AccordionGroup>

## FAQ

<AccordionGroup>
  <Accordion title="Does Elephas upload my documents to train AI?">
    Elephas does not upload your documents or your library to any AI. Cloud AI receives only the question, recent chat messages and the passages needed for that answer. Elephas does not train models on your content. Whether a cloud provider retains requests or trains on them is set by that provider's terms, which the app names for each route.
  </Accordion>

  <Accordion title="Does Elephas store my questions on its servers?">
    The Elephas AI gateway does not store your prompts or answers. Your chats are stored on your computer, and on your other devices if you use [Personal Sync](/docs/privacy-and-security/sync-encryption), where they are end-to-end encrypted.
  </Accordion>

  <Accordion title="Is Elephas AI end-to-end encrypted?">
    No. Requests travel over encrypted connections, but the gateway and the model provider must read your question and passages to answer them. Use Local AI if that is not acceptable for a document.
  </Accordion>

  <Accordion title="Can I use different modes for different work?">
    The mode is a single setting in **Settings › AI**, and you can switch it at any time. A common setup: Local AI for client files, Elephas AI for everything else.
  </Accordion>
</AccordionGroup>

## Related articles

<CardGroup cols={2}>
  <Card title="Choose how Elephas answers" href="/docs/ai-options/choose-how-elephas-answers" />

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

  <Card title="Use Elephas offline" href="/docs/privacy-and-security/use-elephas-offline" />

  <Card title="Privacy overview" href="/docs/privacy-and-security/privacy-overview" />
</CardGroup>


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