v2.2.1

[clusterzx]

What's Changed

You can now download the Chrome extension zip from the repository to start chatting with your documents right out of paperless-ngx itself.
The official extension in chrome store is currently under review for approval.

• Added Chrome Exstension requirements.
• add by @clusterzx in add #19
• Merge pull request add #19 from clusterzx/main by @clusterzx in Merge pull request #19 from clusterzx/main #164
• Open chat from Manual by @Nikorag in Open chat from Manual #154

Full Changelog: v2.2.0...v2.2.1

---

This discussion was created from the release v2.2.1.

[clusterzx]

We have an chrome extension now to work right out of paperless-ngx itself for Document Chat!

Download here: https://github.com/clusterzx/paperless-ai/blob/main/paperless-ai-chrome-extension.zip

Importing ZIP Packages into Chrome for Extensions
Prepare the ZIP: Ensure the ZIP contains the unpacked extension files (e.g., manifest.json).

Enable Developer Mode:

Open Chrome.
Go to chrome://extensions/.
Toggle Developer Mode (top-right corner).
Extract ZIP:

Extract the ZIP to a folder on your computer.
Load the Extension:

Click "Load unpacked" (top-left corner).
Select the folder where you extracted the ZIP.
Test the Extension:

The extension will now appear in the list.
Verify it is working or troubleshoot errors shown in the console.
Done! 🎉

[mamema]

is this new?
[WARN] Invalid date format: 11-2022, using fallback date: 01.01.1990 WARN] Invalid date format: 11-2022, using fallback date: 01.01.1990

where is this 1990 date coming from? Do i have this in the documents? I doubt that.

[mamema]

it's not only the date problem, and it's not your fault, but you have to solve it. :-)
No, i'm just struggling to find a predictable solution, which is not openai.
This seems impossible, tommorow i will post a overall model test (to help others), perhaps then we can talk about solving the problems the models bring into the mix.

[mamema]

hey, young man. :-) are you sure with this:

You are right old man (just joking) 😆 Not 1990, it's 1970 (UNIX Timestamp). Dont know why I had the 90s in my mind. Maybe >because I was born in 90s :D "

because i've further tested this and the document in question, had 01.01.1990 as date. I've changed it to 01.11.2022 in paperless-ngx and let paperless-ai run again. After the run, the document had the date 01.01.1190 again.
So i guess there is some "set date to 01.01.1990" in your code

[mamema]

No absolutly not because of:

1. The AI has extracted in your example 11-2022 (that lacks the day)

2. The AI will likely not produce a better outcome when asking again.

3. Paperless-NGX API only accepts valid date formats xx-xx-xxxx not xx-xxxx.

4. Either way if you just append a 01- to all xx-xxxx dates it still be a not accurate and document valid date. Then a fallback to 01.01.1990 what is a common fallback date in IT is much better to filter out.

another proposed idea. Your setting "use existing tags" (set to yes), i do not know how to use this in a good way, because at the moment (before letting paperless-ai work) i have 409 tags. feeding these into the AI lead to truncated prompts, as this is quite too much!
But why not use this as postprocessing task. still do not feed the tags to the AI to prevent truncated prompts, but evaluate what is coming back and do the "AI work" without AI (semantic compare, similarity, weight) etc.?

[clusterzx]

hey, young man. :-) are you sure with this:

You are right old man (just joking) 😆 Not 1990, it's 1970 (UNIX Timestamp). Dont know why I had the 90s in my mind. Maybe >because I was born in 90s :D "

because i've further tested this and the document in question, had 01.01.1990 as date. I've changed it to 01.11.2022 in paperless-ngx and let paperless-ai run again. After the run, the document had the date 01.01.1190 again.

So i guess there is some "set date to 01.01.1990" in your code

No no you misunderstood me there. I wanted to say that you are right with 1970 and I did set a wrong fallback date there with 1990.

[mamema]

ah, okay then. Another bug found.