feat: Display RAG results

[kevin-on]

Allow users to see the sources contributing to the model's output.

[3-ark]

Yes, also showing the semantic search results like in plug-in "smart connection" and in other rag apps, say MSTY, is very useful too.

[glowingjade]

@3-ark Do you mean something like this? https://github.com/brianpetro/obsidian-smart-connections?tab=readme-ov-file#smart-chat-ai-conversations-based-on-your-notes

[3-ark]

@3-ark Do you mean something like this? https://github.com/brianpetro/obsidian-smart-connections?tab=readme-ov-file#smart-chat-ai-conversations-based-on-your-notes

Yes, the results will give a number, like "0.79", "0.61" to indicate the similarity. My screenshots are more clear. The first one shows the similarity between the current opening note with others in the vault. The second shows the semantic search function of that plug-in. Msty has a similar function. I find it much better than the "outgoing links".

[glowingjade]

@3-ark So the key difference is showing similarity scores, right?

[3-ark]

Difference? If you are referring to the sources (searching results), then, yes. It helps to see the relevance, sometimes model shows irrelevant result in low similarity. But the basic idea of showing the semantic searching results is the core, it let me find all the relevant notes in minutes, for me it's more useful than the generated answer itself.

[glowingjade]

@3-ark We have just added a feature to display search results with a similarity score!
Check out the updates here and here.
The release is scheduled for today.

[3-ark]

Man, it would be another killer! Can't wait to use.

[glowingjade]

It has been released as version 1.0.6!

[3-ark]

I toyed with it a little bit, looks promising. I will get some feedback after using it more later for actual note-taking.

[3-ark]

Thanks for your great work!

[3-ark]

So far so good for me. I just tried and make a new exercise plan using new implemented Gemini 1.5 pro. my only concern is the available model list, I don't know if it could be made in a flexible way. I mean, API providers have more nodes, maybe you guys want to make the list clean but it's a little catching for you when they have new models. But if it can fetch the list automatically, then you don't have to worry about this issue anymore. The only trouble would be too many models in the list. Maybe we can solve it by whitelist in setting.

Another issue is not for me, I just want to provide one possible useful feedback. The rag searching results, now you set it to 15 in default and unchangeable. I am totally happy with this, but maybe someone wants to make it less for a better coherence (because sometimes the similarities are too low for over 10 items) or more (to check more related notes in general, or maybe this could be done by a separated searching function)

Anyway, working perfectly. I love the editing function, it's so smooth.

[3-ark]

Ah, forget to mention. Maybe when you have more users, they will ask for some popular API providers. You guys can check Jan.Ai if you want to add. Those guys are doing a great job in this part.

[glowingjade]

Thanks for your suggestions!

1. We'll look into a better way to handle the model list.
2. For RAG search results, you can actually customize the limit in the Advanced Configuration settings: https://github.com/glowingjade/obsidian-smart-composer/wiki/1.4-Advanced-Configuration#rag-settings
3. Regarding Jan.ai - while I haven't checked if they support OpenAI-compatible APIs, you can use any provider that does through our custom OpenAI-compatible models option: https://github.com/glowingjade/obsidian-smart-composer/wiki/1.4-Advanced-Configuration#custom-openai-compatible-models

[3-ark]

Thanks for your suggestions!

1. We'll look into a better way to handle the model list.
2. For RAG search results, you can actually customize the limit in the Advanced Configuration settings: https://github.com/glowingjade/obsidian-smart-composer/wiki/1.4-Advanced-Configuration#rag-settings
3. Regarding Jan.ai - while I haven't checked if they support OpenAI-compatible APIs, you can use any provider that does through our custom OpenAI-compatible models option: https://github.com/glowingjade/obsidian-smart-composer/wiki/1.4-Advanced-Configuration#custom-openai-compatible-models

It seems my mind was still in the last update. So, I guess it's perfect! Thanks.