DT4 - AI Chat - Unknown error

Now that I’ve installed DT4 Beta, I’d like to test the AI ​​functionality. Unfortunately, I get an “Unknown error” for every question I ask. It doesn’t matter whether I’m asking a general question about a city’s population or a database-related question (e.g., “Show me billings from Vodafone in 2024”). I’ve tried the question in German and English. I probably have a wrong setting. I’m using GPT4All with the “EM German Mistral” model. Does anyone know what’s wrong?
(with Google Translate)

GPT4All Settings I

GPT4All Settings II

DT4 Settings

DT4 Chat Window

Many thanks

Last time I checked this model (probably in 2023) it was a very poor one, I would recommend to try a different one. Or use LM Studio or Ollama with a modern model like Mistral Small 3.1 or Gemma 3.

Or use a commercial model :slight_smile:

@cgrunenberg

In the GPT4All program, I get great results with EM German Mingral.

I’ll test one or two other models.

When I ask “Show me billings from Vodafone in 2024,” I get the following result with Wizard 1.2:

“The Great Gatsby” by F. Scott Fitzgerald: This classic novel is a poignant exploration of the American Dream, love, and wealth in the Jazz Age. It’s set against the backdrop of…"

Under Snoozy, I get the error message “Unknow Error” again.

I give up :smirking_face:

Greetings

GPT4All doesn’t support tool calls, this isn’t even possible using other models. I’d highly recommend Ollama or LM Studio.

1 Like

Okay, then I’ll take a look at the two alternatives mentioned in the next few days. I asked the question directly in the DT4 chat window, not in GPT4All. I used GPT4All because it is mentioned as compatible on their blog.

Compatible with… DEVONthink 4?

Sorry, not mentioned on the blog, but in a post by @cgrunenberg.

I’m now going to test LM Studio

It’s working very well with LM Studio :grinning_face: ! Thanks for your help :waving_hand:

The next beta will work around some template issues of GPT4All when using old models like Mini Orca (Small) or EM German Mistral. In case of other models, e.g. Wizard v1.2, that’s not possible. Models based on LLama or DeepSeek work as expected already and are highly recommended instead.

1 Like

Wow, the initial tests with the DeepSeek model delivered great results :grinning_face:

I downloaded the DeepSeek-R1-Distill-Qwen-7B-Q4_K_M model, or do you recommend another DeepSeek model for DT4?

Glad to hear it but great in what way?

Model recommendations are governed by the limited by your hardware. You’re not going to run a 32b model on anything less than a 32GB RAM machine, and even then it’s likely not going to be very performant.

In the view, for example. With DeepSeek, I get all the documents I’m looking for displayed in the DT document list. With other models, I only see a few links to the documents I’m looking for in the chat window.

Sorry for my bad english.

No worries! Can you post a screenshot of your results,