[Feature Request] Only use Existing Correspondents

[CreekDuzz]

Is your feature request related to a problem? Please describe.
If the AI is free to pick the correspondent, in some cases it will make small variations of the correspondent's name. E.g. [Apple] vs [Apple Inc]. This makes grouping the documents later much more difficult.

Describe the solution you'd like
Like with the existing tags function, limit the AI to only use and consider the existing correspondents. If it cannot be found, allow for the option to add it to the title.

Describe alternatives you've considered
I have tested with sending a full list of the correspondents with the Prompt. It seems to help but it is not a full match.

[clusterzx]

Will implement this till the weekend including existing Tags.

But bear in mind:
If you have thousands of correspondents and thousands of tags then the cost will explode tremendously.

[CreekDuzz]

Yes, you are right. I expect it will not reach those numbers in many cases. I have about 2700 documents collected since 2004 and have 90 correspondents. Yes, I did not include some of the one-off correspondents.

Nevertheless, I would pick higher quality output over a bit more cost.

Thank you!

[thorschtn]

@CreekDuzz

You can already significantly improve the generation of corredpondents by expanding the prompt:

When generating the corredpondent, always create the shortest possible form of the company name (e.g. "Amazon" instead of "Amazon EU SARL, German branch", [add further examples]).

This already improves the accuracy immensely.

Alternatively, you can specify the complete list of all allowed cortrdpondents in the prompt.

[clusterzx]

Thats a great tip! @thorschtn

[Lumpybd]

I'll second this. My list of correspondents is very carefully curated. Letting OpenAI loose on creating new correspondents has created quite a large mess, duplicates etc. I'd prefer to be able to restrict Paperless AI to only select from existing correspondents when populating this field. Perhaps allow for adding new or suggested correspondents when manually processing files?