Unable to get Smart Chats working in Obsidian

[jaidHope]

Hi!
I'm a longtime user of Obsidian and discovered this amazing plugin. However, the plugin keeps hanging/thinking/I don't know what and nothing happens. This behaviour happens when I use remote models like Gemini, as well as local models (eg. local Deepseek running on Ollama). I have attached screenshots of the issue and my settings. Where are the logs for me to inspect what's going wrong? What have I done wrong?
Thanks so much for any assistance.

[brianpetro]

Hi @jaidHope

Thanks for the screenshots and it's nothing you're doing wrong.

That particular error looks like the local model may not currently be compatible with the tool usage required by the lookup.

The Gemini models should work with the tools.

For either, if you open up the developer console logs (CTRL+SHIFT+I on windows) and screenshot the error that appears there when this happens that would help me better understand the specific issue.

Lastly, sometimes the "Send Tool Output in user message" setting fixes things.

Thanks for your patience as we get these issues worked out 🌴

[jaidHope]

obsidian.md-1743878045346.log

[brianpetro]

Thanks for the follow-up and suggestion 😊

It is possible that the DeepSeek model is simply incompatible at the moment.

Since you have dev experience:

The chat models work is through an adapter pattern in the smart-chat-model module, here's a link to the current adapters https://github.com/brianpetro/jsbrains/tree/main/smart-chat-model/adapters

Creating and extending adapters allows flexibility to work with virtually any platform/provider. An adapter essentially converts the request to the model from the incoming OpenAI format (default used by the system) to the necessary format for utilizing the model, then once again converts the response back into the OpenAI format.

One problem I see a lot is that Ollama models aren't working, and that afaik is because different models being served my Ollama may require different format/parameters than what is provided in the current adapter. My hope is that developers using Ollama extend the current class to support more models. This could be as simple as adding a parameter, but also could provide a transformation that converts utilities like tool-calls or structured output into something that can be handled by models without those features.

It's been a while since I said this part here on GitHub, but basically, my primary focus is on the model-agnostic things while leaving space for contributors to integrate with specific platforms/models. For example, that's why making the architecture amenable with the adapter pattern (something that can be seen throughout the JSBrains repository) was, what I believe to be, an important design decision that will ensure the longevity of the project.

That said, I do regularly make fixes to bugs in platform/model-specific implementations because I know for various reasons the project hasn't been in a place that makes it easy for developers to contribute yet.

Unfortunately I didn't see anything in the console logs (PS- screenshot is much preferred in the future) that would lead me to recognize one of those bugs that could be quick fixed.

Lastly, a new Smart Chat plugin is currently in testing with project supporters and will be released publicly/open-source soon. One of my goals with the new release is to further demonstrate the capabilities/simplicities of using the Smart Environment (JSBrains) architecture, and it's nearly a page-one re-write of the current Smart Chat that's much easier to read through and understand how it works. Not only is my hope that it inspires contributors, but also to inspire others to build on the architecture which has been designed from the beginning with an AI-first perspective.

If you made it this far, thanks for reading 😊🌴

[jaidHope]

Hi!
Thanks for the comprehensive explanation. I'm fully on-board with focus toward more model-agnostic work, I agree that it makes better sense from a long-term perspective.
When I get a chance I will tinker a bit with the adapters, but given that you say a new Smart Chat plugin is in testing and it's almost a full rewrite, I won't focus too heavily on it if I struggle to get it working.
I'm def going to be on the lookout for the new plugin's release, eager to give it a spin!
Thanks for the heads-up re: screenshots :)

[neutrinoquark]

Which models do work? Even without local ai the plugin is nice.

[brianpetro]

@neutrinoquark try Qwen-2.5-7b in Ollama. More will be compatible soon (working on it this morning) 🌴