Error creating correspondent

[MichiFr]

I don't know why I'm getting this error in PP-gpt:

time="2026-10-08T08:34:03Z" level=warning msg="Document 40 update failed; \"paperless-gpt-auto\" tag removed and \"paperless-gpt-failed\" tag applied to break the processing loop." document_id=
40
time="2026-10-08T08:34:03Z" level=error msg="Error in background tagging: error in processAutoTagDocuments: error updating document 40: error creating correspondent 'MONTANA Energieversorgung':
 error creating correspondent: 400, {\"name\":[\"correspondent with this name already exists.\"]}"

Of course this correspondent alredy exist from previous OCR runnings and applying the existing correspondent did not fail previously, so why out of a sudden, am I getting this error now instead of just using the existing one?

The spelling is slightly different though, GPT tries to create MONTANA Energieversorgung, however, the existing correspondent reads: Montana Energieversorgung. May this be the source of the error?

[icereed]

🤖 Automated reply from Claude Code, working on behalf of the maintainer.

Hi Michael, yes, the different spelling is exactly the cause, and it's a bug in paperless-gpt, not in your setup.

What happens: paperless-ngx treats correspondent names as unique regardless of case, so MONTANA Energieversorgung and Montana Energieversorgung count as the same name. paperless-gpt, however, looked up existing correspondents by the exact spelling. When the model returned the name in capitals, probably copied from the letterhead, paperless-gpt didn't find your existing correspondent and tried to create a new one, which paperless-ngx rejected. Worse, that error failed the whole update, so the title and tags for document 40 weren't applied either and it got the fail tag. It worked before because the model happened to return the name in the same spelling as your existing correspondent; the model's spelling varies between documents and runs.

The fix is in #1147:

• Correspondents and document types are matched regardless of case, the same way tags already were.
• If a correspondent still can't be used, the other suggestions are applied anyway and the document is only flagged for review.

Once it's merged it will be in icereed/paperless-gpt:unreleased and in the next release.

Until then, for document 40: remove paperless-gpt-failed and add paperless-gpt-auto again. If the model returns the capitalised spelling again, you can also process it once via manual review and pick the existing correspondent.

[MichiFr]

Thanks for confirmation, although answered by Claude ;-)
Nevertheless, thanks for applying a fix for this…👍

[icereed]

Hey Michi, it’s me anyway 😂 but with the amount of PRs and Issues without a lot of time I need to resort to my little helper to let contributors know that I’m working on it 🤓

[MichiFr]

Kein Problem, Hauptsache es wird an einem Fix gearbeitet 😄 Dankeschön!

[icereed]

🤖 Automated reply from Claude Code, working on behalf of the maintainer.

Update: #1147 is merged. In about 15 minutes, once the build finishes, icereed/paperless-gpt:unreleased will:

• match correspondents and document types regardless of case,
• apply the other suggestions even if a correspondent can't be used, and flag the document for review instead.

It will be in the next release too. Feedback from your instance is very welcome.