[BUG]

[wisepass2000]

🐛 Bug Description

When using a custom Ollama model in AutoSort+, the selected model continues to work correctly after restarting Thunderbird, but the Settings UI does not reliably reflect which model is actually in use. After reopening the settings, the custom model field may be blank or the model dropdown may show a default model (e.g. “Llama 2”) that is not installed, creating confusion about whether AutoSort+ has reverted to a default model or is still using the previously configured custom model.

📋 Steps to Reproduce

Open AutoSort+ Settings

Go to Local Ollama Configuration

Set Ollama Model to Custom (enter below)

Enter a valid installed model (e.g. qwen2.5:7b-instruct)

Click Save settings

Restart Thunderbird

Confirm AutoSort+ continues sorting emails successfully

Re-open AutoSort+ Settings

Observe that:

The custom model field is blank or

The model dropdown shows “Llama 2”, even though it is not installed

A warning may appear stating Model “llama2” not installed

❌ Expected Behavior

The Settings UI should clearly and accurately display the actual model currently in use

If a custom model is configured:

The custom model name should remain visible after restart

The model dropdown should not default to an uninstalled model

It should be immediately clear to the user whether:

A custom model is active

Or a fallback/default model is being used

👀 Actual Behavior

AutoSort+ continues to function and sort emails correctly

The Settings UI:

Clears the custom model field or

Displays a default model (e.g. “Llama 2”) that is not installed

This makes it unclear whether AutoSort+:

Is still using the custom model

Or has silently reverted to a default model

📸 Console Output

Browser Console Log (Ctrl+Shift+J in Thunderbird):

[No explicit Ollama errors observed during sorting]
[Warning shown in UI: Model "llama2" not installed]

🔧 Environment Details
Thunderbird

Version: [please fill in]

OS: Windows

OS Version: [e.g. Windows 11]

Ollama Setup

Ollama Version: [output of ollama --version]

Model Used: qwen2.5:7b-instruct

Running on: GPU (NVIDIA)

GPU Model: NVIDIA GPU with 12GB VRAM

Memory Available: 12GB VRAM

AutoSort+

Extension Version: autosortplus-1.2.3.3

Install Method: XPI

✅ Debugging Checklist

Ollama is running (curl http://localhost:11434/api/tags works)

Custom model installed and listed

Test connection passes

Sorting continues to work after restart

UI shows incorrect / misleading model state

Console shows clear indication of active model (not currently visible)

🎯 Additional Context

The issue occurs consistently

Appears limited to settings persistence and UI state

Functionality suggests the custom model remains active internally

The UI presentation makes it unclear which model is actually being used

📌 For Developers

This may be related to:

Model dropdown defaulting on settings reload

Custom model value not being rehydrated into the UI

Lack of a single “source of truth” for active model display

A possible UX improvement would be to:

Always display the active model name (including custom models)

Prevent fallback display of uninstalled default models

When first saving custom model in settings

After Restarting Thunderbird

[wisepass2000]

Thanks again for AutoSort+ — it’s working really well overall, especially with local Ollama models. The sorting performance has been excellent once configured, so hopefully the above is useful rather than nit-picky 🙂

This might be specific to ollama

One small usage note that might help others: because AutoSort+ uses the folder name itself as the prompt, I found accuracy improved a lot by:

Creating very intent-specific folders (e.g. Delivery vs Payments vs Social Media vs Subscriptions)

Using hyphen-separated keywords instead of commas, as commas seemed to sometimes truncate or weaken the prompt (or errors finding a folder in some email type)

Avoiding overlapping concepts in a single folder / AI prompt

Example folder / AI prompt in settings AND adding each folder for email account-

For Delivery & Tracking
Delivery - tracking - shipped - dispatched - out for delivery - delivered - courier - Royal Mail - DPD - Evri

For Payments & Receipts
payment received - receipt - invoice - charged - billing - transaction - amount paid

For Social Media
Facebook - Instagram - LinkedIn - mentioned you - comment - liked your post - page activity

Once the folder names were treated as “mini prompts” rather than labels, the results became very consistent. sorting into the correct folder over 80% of the time with qwen 2.5 instruct as above, I suspect what ever model is used will make a big difference though.

And NO issues with tickets! Since its all done locally.

Thanks again for the work on this — really nice extension, and happy to help test if useful.

[Nigel1992]

✅ Fixed in v1.2.3.4

Hi @wisepass2000! Thanks for the detailed bug report - this has been fixed in version 1.2.3.4.

🔧 What Was Fixed

The issue was caused by how custom Ollama model names were being saved and restored:

1. Custom Model Persistence: When you entered a custom model like qwen2.5:7b-instruct, the extension saved the actual model name but didn't properly restore the UI state. Now the dropdown correctly shows "Custom" and your model name is populated when reopening settings.

2. Auth Token Not Loading: The ollamaAuthToken was being saved but not loaded back into the UI on restart.

3. Missing Settings: AI temperature and rule fallback preferences weren't persisting between sessions.

📥 Download & Install

Download autosortplus-1.2.3.4.xpi

Installation:

1. Download the XPI file
2. In Thunderbird: Menu → Add-ons and Themes
3. Click the gear icon → Install Add-on From File
4. Select the downloaded XPI

✅ What Now Works

After installing v1.2.3.4:

• ✅ Custom Ollama models persist correctly
• ✅ Settings UI accurately shows the active model
• ✅ Auth token is remembered
• ✅ AI temperature setting persists
• ✅ Rule fallback preference persists
• ✅ No more confusion about which model is in use

🧪 Testing Request

Could you please test v1.2.3.4 and confirm that the settings persistence is working correctly for you?

What to test:

1. Install v1.2.3.4 from the release above
2. Configure your custom Ollama model (qwen2.5:7b-instruct)
3. Set your auth token (if applicable)
4. Adjust AI temperature if desired
5. Save settings
6. Restart Thunderbird
7. Open AutoSort+ settings again
8. Confirm all settings are preserved and displayed correctly

Please let me know if everything works as expected or if you encounter any issues. Once confirmed working, I'll merge these changes into the main branch as this fix is stable and ready for production use. Thanks! 🙏

---

Thanks again for the detailed report and the helpful suggestions about using folder names as prompts. That's excellent advice for other users! 🙌

[Nigel1992]

Closing as fixed in v1.2.3.4. Please reopen if you encounter any issues with the fix!

[wisepass2000]

I tested new release and saving settings persists for the custom model thanks for sorting.