> ## 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.

# Run AI Locally on Your Computer with Elephas

> Run a local AI model on your Mac or PC with Elephas. Download a model once, then ask your documents without sending your questions or files to an AI provider.

With Local AI, a language model runs on your own computer and writes the answers. Your questions, your documents and your chats are not sent to any AI provider. You download a model once, from about 1.2 GB to 5.3 GB, and after that Elephas can answer with your Wi-Fi off, as long as it has confirmed your plan recently.

<Info>
  Local AI is included with the Elephas trial and with Personal and Professional plans. In a Team workspace, it works if the Team's AI policy allows it. See [Plans and pricing](/docs/plans-and-billing/plans-and-pricing).
</Info>

## Before you start

* **Memory.** The smallest model runs on most computers. Bigger models need more memory. See [Choose a local model](/docs/ai-options/choose-a-local-model).
* **Disk space.** Enough free space for the model you pick, plus a small AI engine.
* **A connection for the download**, and to confirm your plan before you go offline.
* **Sign in**, with an active trial or plan.

## Steps

<Steps>
  <Step title="Open Settings › AI">
    Press ⌘, on Mac or Ctrl+, on Windows, then click **AI**.
  </Step>

  <Step title="Choose Local (on this computer)">
    Under **How Elephas answers**, click **Local (on this computer)**. Elephas picks the model that suits your computer and asks before downloading, for example **Download Qwen 4B, 2.6 GB?**

    To pick a different model first, click **Not now**, open **Which model**, choose one, then click **Local (on this computer)** again. See [Choose a local model](/docs/ai-options/choose-a-local-model).
  </Step>

  <Step title="Download the model">
    Click **Download**. The card shows **Downloading…** with how much is done and a progress bar. You can keep reading your Brains and chats, but Ask can't answer with Local AI until the model is ready.

    * Click **Cancel** to pause. The part already downloaded is kept, the card shows **Download paused.**, and **Resume** carries on later. Your previous AI option is used in the meantime.
    * If your connection drops, the card says **Waiting for network. The download continues by itself.**
    * If you quit Elephas, the download picks up where it left off the next time you open it.
  </Step>

  <Step title="Wait for Local AI to start">
    After the download, Elephas checks the files (**Checking the files…**) and starts the model (**Starting local AI…**). The first start can take a minute. When it reads **Local AI ready** followed by the model name, you're set.
  </Step>

  <Step title="Ask a question">
    The chat bar reads **Answered by: This computer** and the model name, for example **Answered by: This computer · Qwen 4B**.
  </Step>
</Steps>

The steps are the same on Mac and Windows.

## How it works

### What stays on your computer

The model runs on your computer through a local engine that only accepts connections from your own computer. When you ask, Elephas searches your documents locally and hands the passages to the local model. Your question, the passages, your chat history and the answer are not sent to any AI provider. In the app's words: "Answers are generated on this computer. Nothing is sent anywhere." See [What data is sent to AI](/docs/privacy-and-security/what-data-is-sent-to-ai) to compare with the cloud options.

### The plan check

Local AI needs an Elephas trial or plan, and Elephas confirms it online. A passed check lasts up to 60 minutes and is cleared when you restart Elephas. The check sends only a device ID, never a question, document or answer. If the check is due and Elephas can't reach its service, Ask says **Elephas could not confirm your plan right now. Check your connection and try again.** Reconnect briefly and ask again.

So Local AI works with Wi-Fi off for up to an hour after a check, not indefinitely. For everything else that still goes online, see [Use Elephas offline](/docs/privacy-and-security/use-elephas-offline).

### The models

Elephas offers three Qwen models: **Qwen 2B**, **Qwen 4B** and **Qwen 9B**. **Choose for me**, the default, picks the best one your computer can run comfortably. Bigger models give better answers but need more memory and run slower. See [Choose a local model](/docs/ai-options/choose-a-local-model).

Every downloaded file is checked against a fixed fingerprint before Elephas uses it. A damaged download is fetched again.

### What to expect from answers

* **Slower than cloud AI.** If no words appear for a while, Ask says **Thinking… this can take a minute on this computer.**
* **One pass.** The local model answers from the passages Elephas found. It doesn't read further into your documents on its own, so focused questions over a few documents work best. See [Tips for better answers](/docs/ask/tips-for-better-answers).
* **Citations still work.** Answers cite your documents like any other answer.
* **Never a silent switch to the cloud.** While Local AI is downloading or starting, Send stays off. Elephas never falls back to a cloud AI without you choosing it.

### Switching away and back

Choosing another option stops the local engine and frees its memory. The model stays on your computer, so switching back to Local AI starts it again without a download.

### Personal information protection

Protection is for cloud AI. Since nothing is sent to an AI provider in Local AI mode, it isn't needed. See [Personal information protection](/docs/privacy-and-security/personal-information-protection).

### Team workspaces

Local AI also answers questions about Team Brains, if the Team Owner's AI policy allows it. Elephas remembers the last policy it saw for each Team, so this works offline too. If the policy turns cloud AI off, Local AI is refused as well. See [Team AI policy](/docs/teams/team-ai-policy).

## Troubleshooting

<AccordionGroup>
  <Accordion title="Local (on this computer) is greyed out">
    It says **Not enough free memory on this computer.** Your computer can't run even the smallest model. Close other apps and reopen Settings, or use Elephas AI or your own key. See [Local AI troubleshooting](/docs/troubleshooting/local-ai-troubleshooting).
  </Accordion>

  <Accordion title="Not enough free disk space">
    The card says how much space to free, for example **Not enough free disk space. Free up 1.4 GB and try again.** Free that space, then click **Retry**.
  </Accordion>

  <Accordion title="Local AI couldn't start, or stopped unexpectedly">
    Click **Try again**. If it keeps failing, click **Set up again** to check and repair the files, or pick a smaller model. See [Local AI troubleshooting](/docs/troubleshooting/local-ai-troubleshooting).
  </Accordion>

  <Accordion title="Too much context for the local model">
    The question and its passages don't fit in the local model. Ask a shorter question or choose fewer Brains, or narrow it to specific documents.
  </Accordion>
</AccordionGroup>

## FAQ

<AccordionGroup>
  <Accordion title="Does Local AI cost extra?">
    No. It is included with your trial or plan, and it doesn't use your Elephas AI allowance.
  </Accordion>

  <Accordion title="Where does the model come from?">
    Elephas downloads the Qwen models from Hugging Face, pinned to a fixed version, with Elephas's own server as a fallback. The engine that runs them comes from Elephas. Every file is checked against a fixed fingerprint.
  </Accordion>

  <Accordion title="Can I use Local AI with no internet at all?">
    Not indefinitely. Elephas needs to confirm your plan at least once an hour, and after every restart. Between checks, Local AI works with no connection.
  </Accordion>
</AccordionGroup>

## Related articles

<CardGroup cols={2}>
  <Card title="Choose a local model" href="/docs/ai-options/choose-a-local-model" />

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

  <Card title="Choose how Elephas answers" href="/docs/ai-options/choose-how-elephas-answers" />

  <Card title="Local AI troubleshooting" href="/docs/troubleshooting/local-ai-troubleshooting" />
</CardGroup>


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