Troubleshooting

What it is:

This is expected. 4.0 no longer asks for access to every website. Instead the browser asks for each server address the first time you save a provider that uses it, and, when attaching a page cannot use your click's one-off access, once for all sites, after a dialog explaining why.

Solution:

1. Click Allow. The provider is saved and requests go through.
2. If you dismissed the prompt, save the provider again to be asked again.
3. To review or revoke grants later, open chrome://extensions, click Details on EZGPT.APP and look under Site access.

Solution:

1. A new host or port needs its own grant. Save the provider again and allow the new address.
2. localhost and 127.0.0.1 count as different addresses. Use the one you granted, or grant both.

Solution:

1. In LM Studio open Developer and make sure the server is started. Load a model first.
2. The preset URL is http://localhost:1234/v1. If you changed LM Studio's port, change it here too.
3. Make sure you allowed access to localhost:1234 when asked. See Permission prompts.
4. Click Test connection in the provider settings.
5. For LM Studio on another machine, tick "Serve on local network" in LM Studio and use that machine's IP in the URL.

Solution:

1. Setting OLLAMA_ORIGINS is no longer needed in 4.0. Once you allow the origin permission for localhost:11434, the request is made with extension privileges and Ollama's CORS check does not apply. Remove the old variable if you like; it does no harm.
2. Check Ollama is running: open http://localhost:11434 in a tab. It should say "Ollama is running".
3. The preset URL is http://localhost:11434/v1.
4. Pull at least one model, for example ollama pull llama3.2, then click Discover models.
5. For Ollama on another machine, start it with OLLAMA_HOST=0.0.0.0 on that machine and use its IP in the URL.

Solution:

1. LM Studio lists only models that are loaded or downloaded. Load one.
2. Ollama lists pulled models. Pull one.
3. Some custom servers do not implement the models endpoint. Edit the provider and type the model name into Default model.

Solution:

1. Check the key is pasted in full with no spaces, and that it belongs to the service the preset is for.
2. Check the account has credit or an active plan.
3. Make sure you allowed access to the service's address when saving. See Permission prompts.
4. Click Test connection to see the server's own error text.

Solution:

1. Edit the provider and click Discover models again to refresh the list. Cloud catalogues change.
2. On OpenRouter some models need a paid balance or a specific provider setting on your account.

What it is:

Browsers do not let extensions read their own internal pages, so Attach page and the context menu cannot work on chrome:// and edge:// pages, the Chrome Web Store, the new-tab page, or the built-in PDF viewer.

Solution:

1. Switch to a normal web page and try again.
2. For a PDF, select the text you need and paste it into the message box.
3. On a normal page that still fails, allow access to the site when the browser asks, then try again.

Solution:

1. Select the text on the page first, then click Attach page or press Ctrl+Shift+P. With no selection, the readable page text is attached.
2. Or right-click the selection and choose Ask EZGPT.APP about….

Solution:

1. Reasoning models put their thinking in the collapsible block above the answer. Expand it to see what the model did. If the answer never arrives, the model may have run out of tokens while thinking. Raise Max tokens in the Parameters dialog (sliders icon in the composer) or leave it empty.
2. Check the context window in the provider settings matches your model. A too-large value can make a local server reject the request.

Solution:

1. Use a vision model such as Qwen-VL, LLaVA, Pixtral or GPT-4o. Text-only models cannot see images.
2. In LM Studio and Ollama make sure the vision variant is the one loaded.

Solution:

1. Lower the context window in the provider settings to the model's real limit so older messages are trimmed sooner.
2. Start a new chat with Ctrl+N.
3. Attached page text counts. Attach a selection rather than the whole page.

Solution:

1. EZGPT.APP 4.0 needs Chrome or Edge 116 or newer. Check chrome://version.
2. Try Ctrl+Shift+E. If another extension owns that shortcut, change it at chrome://extensions/shortcuts.
3. Reload the extension from chrome://extensions.

Solution:

1. Click the puzzle-piece menu in the toolbar and pin EZGPT.APP.
2. Check the extension is enabled at chrome://extensions.

They are replaced by the side panel, which stays open while you browse, and by Open in tab for a full-size view. Open the side panel with the toolbar icon or Ctrl+Shift+E.

Solution:

1. Migration runs once, the first time 4.0 starts. Reload the extension from chrome://extensions and open the panel again.
2. Chats are now grouped by day. Open the chat list with ☰ or Ctrl+K, and tick Filter: current provider to see only one provider's chats.
3. If you exported a backup from 3.x, use Settings → Data → Import backup.
4. Still missing? Send a report with your browser version and what was there before.

In 4.0 Enter sends and Shift+Enter adds a line.