Skip to main content

Dan-EL models

Dan-EL is Glitr’s on-device assistant. Chat answers run on this device after you load a model. Network may still be needed for the first model download and for refreshing the docs index.

What you'll need​

  • Enough free storage for a model pack (size depends on the model you pick).
  • On the web: a browser that supports the local inference path Glitr uses (WebLLM).
  • On desktop/mobile native builds: the packaged Candle path when enabled.

Open Dan-EL​

  1. Tap the sparkles bubble on any page, or open the AI / Dan-EL page from the nav.
  2. If no model is loaded, Glitr prompts you to download or select one.

Download a model​

  1. Open Dan-EL.
  2. Open model / storage settings (gear or Models).
  3. Pick a model from the list.
  4. Confirm Download and wait for progress to finish.
  5. Select the downloaded model as active.

Stay on Wi‑Fi for large downloads when you can.

Manage storage​

  1. Open Dan-EL → model / storage settings.
  2. Review how much space models use.
  3. Remove models you no longer need to free space.
  4. Keep at least one model if you want chat answers without re-downloading.

Ask the assistant​

  1. Open the sparkles bubble or the AI page.
  2. Type a question about Glitr (setup, chats, Drive, calls, and so on).
  3. Send. Dan-EL answers using the loaded model and on-device docs index when available.

Page-aware tips appear as a Getting started card when you open the bubble on a new screen — In-app assistant.

Image generation​

When your build includes image generation:

  1. Open Dan-EL.
  2. Choose the image / generate mode if offered.
  3. Describe the image and run generate.
  4. Save or share the result when the UI offers it.

If your build does not list image generation, that feature is not enabled in this package yet.

What you should see​

  • A progress indicator while a model downloads.
  • Chat replies appear in the bubble or AI page thread.
  • Getting-started suggestions match the page you are on.

If something goes wrong​

  • Download stuck: check disk space and network; retry from model settings.
  • No reply: confirm a model is selected and loaded.
  • More help: Troubleshooting and In-app assistant.

Next steps​

For the curious

Web builds typically use WebLLM in the browser; native builds may use Candle. The docs catalog can be refreshed from glitr.io (ai-index.json) while answers themselves stay local after the model is on device.