[BUG]

[B0R1K]

🐛 Bug Description

A clear and concise description of what the bug is.

Analysis with AI throws these errors:

[Ollama] Tab injection chat failed: HTTP 404: Not Found background.js:704:25
analyzeEmailContent moz-extension://937918cb-fbf4-4c04-965a-e5572a735921/background.js:704
Error analyzing email: Error: HTTP 404: Not Found
ollamaChatViaTab moz-extension://937918cb-fbf4-4c04-965a-e5572a735921/background.js:101

📋 Steps to Reproduce

Steps to reproduce the behavior:
right click on any email, select "Analyze with AI", see it error in Browser Console

❌ Expected Behavior

What you expected to happen.
get some labels or see it do something?!

👀 Actual Behavior

What actually happened instead.
Nothing

📸 Console Output

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

<img width="1327" height="429" alt="Image" src="https://github.com/user-attachments/assets/44c3930e-4f4d-4ce5-8649-e25a6425a1e0" />

---

🔧 Environment Details

Thunderbird

• Version: 147.0
• OS: Windows
• OS Version: Windows 11

Ollama Setup

• Ollama Version: 0.15.1
• Model Used: tinyllama
• Running on: CPU / GPU (which GPU model?)
• Memory Available: (e.g., 8GB, 16GB)

AutoSort+

• Extension Version: e.g., 1.2.3.1-ollama-test
• Install Method: XPI

---

✅ Debugging Checklist

• Ollama is running: curl http://localhost:11434/api/tags returns models
• Model installed: ollama list shows your model
• Test connection passes: Settings → Test Connection works
• Thunderbird restarted after AutoSort+ install
• Console logs checked (Ctrl+Shift+J shows [Ollama] messages)
• Email is plaintext (not HTML-only)
• Model responds locally: ollama run tinyllama "test"

---

🔍 Manual Testing Steps

1. Verify Ollama API works:

curl http://localhost:11434/api/tags

Should list your installed models.

2. Test direct API call:

curl -X POST http://localhost:11434/api/chat \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tinyllama",
    "messages": [{"role": "user", "content": "What is email classification?"}],
    "stream": false
  }'

Should return a response from the model.

3. Check model performance:

ollama run tinyllama "Classify this email: [subject line here]"

4. Enable verbose logging:

• Ctrl+Shift+J in Thunderbird
• Analyze an email
• Copy all [Ollama] log entries

---

📝 Error Message (if applicable)

[Paste the full error message here]

🎯 Additional Context

• What were you trying to do?
• Does it happen consistently or randomly?
• Have you tried other models?
• Any recent Thunderbird or Ollama updates?

---

📌 For Developers

How to investigate tab injection issues:

• Check if hidden tab at http://localhost:11434 opens and closes
• Verify window.__ollama_result is populated
• Check network tab for POST to /api/chat
• Inspect returned JSON structure from Ollama API

[B0R1K]

Ran the Diagnostics and its all good!

🔍 OLLAMA DIAGNOSTICS

📋 Test 1: List Models Endpoint
URL: http://localhost:11434/api/tags
Status: 200 OK
✓ SUCCESS - Found 1 models
Installed models: tinyllama:latest

🔢 Test 2: Version Endpoint
URL: http://localhost:11434/api/version
Status: 200 OK
✓ SUCCESS - Ollama version: 0.15.1

⬇️ Test 3: Pull Endpoint Check
URL: http://localhost:11434/api/pull
Note: This endpoint is used for downloading models

==================================================
📊 SUMMARY:

✓ Ollama is running and accessible
✓ API base URL: http://localhost:11434

[B0R1K]

[Nigel1992]

Hi @B0R1K, thanks for the detailed report!

I’ve been developing and testing this extension primarily on Xubuntu Linux, and it looks like we've hit a common snag with how Ollama handles model names and tags across different environments.

The 404 Not Found in your logs indicates that the Ollama server is active, but it doesn't recognize the specific model name/URI being requested. In Ollama, if you don't specify a tag, the API automatically looks for :latest. If you have the model installed under a specific version tag (like tinyllama:1.1b) but not tinyllama:latest, the API will return a 404.

---

🛠️ Diagnostic Checklist (Windows)

To pinpoint exactly why the server is rejecting the request, please run these commands in PowerShell:

1. Exact Name Test

This test forces the :latest tag. If this returns a JSON response instead of a 404, the issue is a tag mismatch:

Invoke-RestMethod -Method Post -Uri "http://127.0.0.1:11434/api/chat" -Body '{"model": "tinyllama:latest", "messages": [{"role": "user", "content": "hi"}], "stream": false}' -ContentType "application/json"

2. API Model List

Check exactly how the API sees your installed models. This will show the registered names and tags that Ollama expects:

Invoke-RestMethod -Uri "http://127.0.0.1:11434/api/tags" | Select-Object -ExpandProperty models | Select-Object name

3. Check the Live Server Logs

If the tests above don't clarify the issue, we can look at the "smoking gun" in the Ollama logs:

1. Open PowerShell and run:

Get-Content "$env:LOCALAPPDATA\Ollama\server.log" -Wait -Tail 20

2. While that is running, try to use the extension in Thunderbird.
3. Paste any new lines that appear here. Look specifically for lines containing HTTP/1.1 404.

---

✅ Quick Fixes to Try

1. Remove Trailing Slashes: In the AutoSort+ settings, ensure your Ollama URL is exactly http://127.0.0.1:11434 with no trailing slash at the end.
2. Use IP instead of Localhost: Using 127.0.0.1 instead of localhost avoids common IPv6 resolution issues on Windows.
3. Create an Alias: if your model is named something like tinyllama:1.1b, you can create a "latest" alias by running this in your terminal:

ollama cp tinyllama:1.1b tinyllama:latest

Please let me know what the logs show or if Test #1 succeeds!

[B0R1K]

hi @Nigel1992

Changing model in UI "tinyllama:latest" resolves 404, but no labels yet
Skipping message because generated label was null/empty background.js:1237:29

Test 1
model : tinyllama:latest
created_at : 2026-01-26T03:36:19.6430998Z
message : @{role=assistant; content=Absolutely! Here's an updated version of the sentence with "you"
replaced by "your":

                   You are an excellent AI assistant.}

done : True
done_reason : stop
total_duration : 541165100
load_duration : 30600100
prompt_eval_count : 35
prompt_eval_duration : 134523100
eval_count : 33
eval_duration : 298679800

Test 2
name

tinyllama:latest

Test 3
PS C:\Users\Borik> Get-Content "$env:LOCALAPPDATA\Ollama\server.log" -Wait -Tail 20
time=2026-01-25T22:31:18.171-05:00 level=WARN source=runner.go:153 msg="truncating input prompt" limit=2048 prompt=7047 keep=5 new=2048
[GIN] 2026/01/25 - 22:31:19 | 200 | 1.4229088s | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:31:20 | 200 | 0s | 127.0.0.1 | GET "/"
time=2026-01-25T22:31:20.702-05:00 level=WARN source=runner.go:153 msg="truncating input prompt" limit=2048 prompt=5453 keep=5 new=2048
[GIN] 2026/01/25 - 22:31:21 | 200 | 814.9521ms | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:31:22 | 200 | 0s | 127.0.0.1 | GET "/"
[GIN] 2026/01/25 - 22:31:23 | 200 | 790.6888ms | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:31:23 | 200 | 0s | 127.0.0.1 | GET "/"
time=2026-01-25T22:31:24.468-05:00 level=WARN source=runner.go:153 msg="truncating input prompt" limit=2048 prompt=14054 keep=5 new=2048
[GIN] 2026/01/25 - 22:31:25 | 200 | 1.5468495s | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:31:26 | 200 | 0s | 127.0.0.1 | GET "/"
time=2026-01-25T22:31:27.095-05:00 level=WARN source=runner.go:153 msg="truncating input prompt" limit=2048 prompt=5398 keep=5 new=2048
[GIN] 2026/01/25 - 22:31:28 | 200 | 1.4071406s | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:31:29 | 200 | 0s | 127.0.0.1 | GET "/"
[GIN] 2026/01/25 - 22:31:31 | 200 | 1.7252573s | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:31:31 | 200 | 0s | 127.0.0.1 | GET "/"
time=2026-01-25T22:31:32.361-05:00 level=WARN source=runner.go:153 msg="truncating input prompt" limit=2048 prompt=5374 keep=5 new=2048
[GIN] 2026/01/25 - 22:31:33 | 200 | 987.0821ms | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:36:19 | 200 | 541.1651ms | 127.0.0.1 | POST "/api/chat"
[GIN] 2026/01/25 - 22:36:43 | 200 | 580.5µs | 127.0.0.1 | GET "/api/tags"

[Nigel1992]

Hi @borik — thanks for testing. Good that switching the UI model to tinyllama:latest fixed the 404.
I pushed a fix (commit 53cc0a9) to main that makes Ollama response parsing and label matching more robust: improved handling for arrays/objects and tolerant label matching.
Please re-run with tinyllama:latest. If you still see "Skipping message because generated label was null/empty", please paste the full AI response JSON or console logs (the Ollama response) here and I'll refine parsing further.
Thanks for the report — I can add extra debug logging if helpful.

Here's the test release: https://github.com/Nigel1992/AutoSort-Plus/releases/tag/v1.2.3-ollama-test-20260126

[B0R1K]

hi @Nigel1992,
I am still getting "Skipped Msg". Where can I see the complete AI response?

[Nigel1992]

Hi @borik — thanks for the follow-up. Sorry it's still skipping — please capture the full AI response so I can refine the parser.

1. Thunderbird Browser Console

• Open: Tools → Developer Tools → Browser Console (or Ctrl+Shift+J).
• Reproduce the "Skipped Msg". Look for these AutoSort+ log lines:
  • No label extracted from response: <-- the full AI response object will follow this line
  • Full API response data: <-- (for non‑Ollama providers) the JSON payload is printed here
  • Skipping message because generated label was null/empty
• Copy/paste the JSON object that appears after "No label extracted from response:" into this issue.

2. Ollama server log (if using local Ollama)

• PowerShell (Windows):

  Get-Content "$env:LOCALAPPDATA\Ollama\server.log" -Wait -Tail 50

• Linux/macOS (example path — adjust if different):

  tail -n 50 -f /path/to/ollama/server.log

3. If you prefer quicker debugging, I can add a temporary debug mode that writes the full AI JSON to a file or shows it in a notification — reply 'add-debug' and I'll push a new prerelease with that.

Thanks — paste the JSON or server logs here and I'll iterate immediately.

[Nigel1992]

Hi @borik — thanks for the follow-up. Sorry it's still skipping — please capture the full AI response so I can refine the parser.

1. Thunderbird Browser Console

• Open: Tools → Developer Tools → Browser Console (or Ctrl+Shift+J).
• Reproduce the "Skipped Msg". Look for these AutoSort+ log lines:
  • No label extracted from response: <-- the full AI response object will follow this line
  • Full API response data: <-- (for non‑Ollama providers) the JSON payload is printed here
  • Skipping message because generated label was null/empty
• Copy/paste the JSON object that appears after "No label extracted from response:" into this issue.

2. Ollama server log (if using local Ollama)

• PowerShell (Windows):

  Get-Content "$env:LOCALAPPDATA\Ollama\server.log" -Wait -Tail 50

• Linux/macOS (example path — adjust if different):

  tail -n 50 -f /path/to/ollama/server.log

3. If you prefer quicker debugging, I can add a temporary debug mode that writes the full AI JSON to a file or shows it in a notification — reply 'add-debug' and I'll push a new prerelease with that.

Thanks — paste the JSON or server logs here and I'll iterate immediately.

[B0R1K]

[Nigel1992]

Just to confirm, did you add labels in settings ?
If so, which ones?

The debugger is saying it thinks "promos" is fitting but you didnt add this yet to your label list.

I should probably rewrite the extension to either 1. create a folder on your IMAP with the new label or 2. only use labels you provide and make it match to the closest match possible.
Which one do you prefer ?

Note that option 1 would potentially make a ton of folders, i'd advice option 2.

[B0R1K]

this is my config, what am i missing

[Nigel1992]

@B0R1K

My AutoSort tool thinks these emails belong in a "Promos" folder. However, since you don't have a "Promos" label set up, the extension is currently ignoring them.

I’m planning to update the extension to fix this. Which approach do you prefer?

1. Auto-Create: The extension creates the missing folder on your email account automatically. (Note: This could clutter your sidebar with many new folders.)
2. Smart Match: The extension only uses your existing labels and finds the "closest match" available.

My recommendation: I suggest Option 2 to keep your inbox organized. Which one would you prefer?

[B0R1K]

Probably Number 2

[Nigel1992]

Alright, I will be updating the extension very soon.
Thanks for your patience :)