Filtering of summary result

[jangrewe]

Describe the feature you'd like

Hi, i'm currently setting up Hoarder will all the bells and whistles (and it is amazing!), but the one think that bugs me is that the "AI" summary always looks like " Hmm, so i should This is the summary.", so it includes the <think></think> tags that Ollama/DeepSeek-R1 returns.

Is there any way to add e.g. a regex to filter out this thinking/reasoning output?

Describe the benefits this would bring to existing Hoarder users

Cleaner and more concise summaries.

Can the goal of this request already be achieved via other means?

No

Have you searched for an existing open/closed issue?

• [x] I have searched for existing issues and none cover my fundamental request

Additional context

No response

[MohamedBassem]

hmmm, this is an interesting one. I don't think a lot of people have been using reasoning models with hoarder. Does this even work for tagging?

[jangrewe]

Oh, absolutely! 😁

[Ishiguro872]

Interesting. How did you get this to work? I recently tried using deepseek-r1 but the tags never work. Not sure if it's because of the thinking portion of the model that just times it out

[jangrewe]

Uhm, i didn't do anything special, i just made sure that it works in Ollama (with Open WebUI) and then set it as the model to be used for text inference in Hoarder. 🙂

[insanemal]

I'm also successfully using R1. It's fantastic for tagging. However there are two issues, both already have tickets.

This one is the most pressing however. It makes the summaries almost useless, especially in terms of formatting.

So being able to filter them out would be fantastic. So would being able to edit the default prompts a bit more.

Just let me know what you need to assist with this, I can get logs/examples.

[huyz]

Out of curiosity, how are reasoning models better? More appropriate tags? Just better choice of words?

[cultpony]

I've encountered this issue with Nebius' API offering, which seems to sometimes go for a thinking model and the tags just end up in the summary, even when I try to pick a model that has no reasoning mode like this. The Tags are fine however. The output modes for structured, plain or json all yield the same result.