> ## 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.

# Local AI Not Working: Memory, Download and Startup Errors

> Fix Local AI problems in Elephas: not enough memory, a paused or damaged model download, a full disk, the local engine failing to start, and slow answers.

Local AI runs a model on your computer, so most problems come from memory, disk space or the one-time model download. Open **Settings › AI** and read the status line under **Local AI**. Each status has one button that fixes it.

<Info>
  Local AI is included with an Elephas trial or paid plan. In Personal workspaces it needs a short online plan check about once an hour and after each restart.
</Info>

## Before you start

* Open **Settings › AI** and check that **Local (on this computer)** is selected under **How Elephas answers**.
* Local AI downloads a model once, about 1.2 GB to 5.3 GB depending on the model, plus the on-device AI engine. Make sure you have a connection and free disk space for the first setup.
* See [choose a local model](/docs/ai-options/choose-a-local-model) for which model suits your computer.

## Status messages in Settings › AI

| Status | Cause | Fix |
| - | - | - |
| **Not enough free memory on this computer.** | Your computer cannot run even the smallest model. | Quit other apps and restart Elephas. If the message stays, use Elephas AI or your own API key on this computer. |
| **Download paused.** | You clicked **Cancel** during the download. | Click **Resume**. Elephas keeps what was already downloaded. |
| **Waiting for network. The download continues by itself.** | You went offline during the download. | Reconnect. The download resumes on its own. |
| **Not enough free disk space. Free up \[size] and try again.** | Your disk does not have room for the model. | Free up at least the amount shown, then click **Retry**. |
| **That download was damaged. Elephas will fetch it again.** | The model file did not match its expected fingerprint, so Elephas deleted it. | Click **Retry**. If it repeats, your network may be altering downloads. See the firewall note below. |
| **The download couldn't finish. Try again.** | The download server did not answer or the connection broke. | Click **Retry**. |
| **Elephas can't write to its data folder.** | Your user account cannot write to the Elephas folder. | Check folder permissions (see [Elephas won't start](/docs/troubleshooting/elephas-wont-start)), then click **Retry**. |
| **Another copy of Elephas is already downloading this.** | Elephas is open twice. | Close the other copy, then click **Retry**. |
| **Local AI couldn't start.** | The local engine failed to start or stopped, often because memory ran short or the model is too large for this computer. | Quit other apps and click **Try again**. If it fails again, choose a smaller model (below). |
| **Local AI needs to be set up again.** | The model file is missing, for example after it was deleted. | Click **Set up**. |
| **A different model suits this computer better.** | A hint, not an error. Your computer can now run a different model. | Click **Set up again** if you want to switch. Nothing changes until you do. |

## Messages in Ask

| Message | Fix |
| - | - |
| **Local AI isn't ready yet. Check Settings.** | The model is still downloading, starting or needs attention. Click **Open Settings** and read the status line. |
| **Local AI stopped unexpectedly. Try again.** | Ask again. If it repeats, quit other apps or choose a smaller model. |
| **Too much context for the local model. Try a shorter question or fewer Brains.** | Local models read less text at once than cloud AI. Ask about fewer Brains, or pick specific documents. |
| **Elephas could not confirm your plan right now. Check your connection and try again.** | Go online briefly so Elephas can check your plan, then ask again. See [use Elephas offline](/docs/privacy-and-security/use-elephas-offline). |
| **Thinking… this can take a minute on this computer.** | Not an error. Local answers start more slowly than cloud ones. Wait, or use a smaller model. |

## Steps: switch to a smaller model

<Steps>
  <Step title="Open the model list">
    Open **Settings › AI**. Under **Local AI**, find **Which model**.
  </Step>

  <Step title="Pick a smaller model">
    Choose a smaller model, or **Choose for me** to let Elephas pick the best model this computer can run comfortably. Models marked **Probably too big for this computer. You can still try it.** are the likeliest to fail.
  </Step>

  <Step title="Download it">
    Changing the choice downloads nothing on its own. Click the download button that names the model and its size, and confirm.
  </Step>

  <Step title="Wait for Local AI ready">
    When the status reads **Local AI ready** with the model name, ask again.
  </Step>
</Steps>

## How it works

* **Which model.** Elephas offers three models. **Choose for me** picks by your computer's memory: Qwen 2B on any computer, Qwen 4B from 8 GB of memory, and Qwen 9B from 20 GB. On Windows, Elephas also checks how much memory is free right now and steps down if needed.
* **Starting up.** The engine starts when you first ask with Local AI and stays loaded until you quit or switch away. If it stops, Elephas restarts it on your next question, a few times, before it shows **Local AI couldn't start.**
* **What stays local.** In Local AI mode, your question, the passages it reads and the answer are generated on your computer and not sent to any AI provider. Elephas still goes online for the plan check. See [what data is sent to AI](/docs/privacy-and-security/what-data-is-sent-to-ai).
* **No silent fallback.** If Local AI is not ready, Elephas does not quietly send your question to a cloud AI. It tells you and waits.

**Firewalls.** The local models download from `huggingface.co` and the on-device AI engine from `next.elephas.ai`. On a work network, ask your IT team to allow both.

## Troubleshooting

<AccordionGroup>
  <Accordion title="Local AI is greyed out">
    Your computer does not have enough memory for the smallest model. On Windows, quit memory-heavy apps and restart Elephas. Otherwise use Elephas AI or your own API key.
  </Accordion>

  <Accordion title="Answers are very slow">
    Local answers depend on your computer. A smaller model answers faster. Open **Details** under Local AI to see load time, time to first word, speed and memory use for the last answer.
  </Accordion>

  <Accordion title="I set up Local AI but answers still say Elephas AI">
    Check the badge under the answer. It reads **Answered by: This computer** for Local AI. If not, make sure **Local (on this computer)** is selected in **Settings › AI**.
  </Accordion>
</AccordionGroup>

## Related articles

<CardGroup cols={2}>
  <Card title="Run Local AI" href="/docs/ai-options/run-local-ai" />

  <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="System requirements" href="/docs/getting-started/system-requirements" />
</CardGroup>


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