Contextual auto-tagging in Obsidian using AI

Feature requests
, , ,
1

Hey Guys!

I recently came across an interesting podcast with our new CEO, Kepano. During the conversation, it became apparent that Kepano was not aware of Napkin’s existence. This surprised me, as I’ve been eagerly anticipating AI integration in Obsidian, especially auto-linking features similar to those in Napkin. However, Kepano announced that Obsidian would NOT venture into AI for another two years.

I use Napkin, an app that leverages AI to link ideas, and it has significantly enhanced my PKM experience. Yet, I still prefer Obsidian for its flexibility. As a result, I supply my data to both Obsidian and Napkin using Readwise sync, relying on Napkin to browse through my Garden until Obsidian introduces its AI. Kepano’s words were disheartening, as I have been blindly dumping data into Obsidian, waiting for the day when AI magically organizes it all at once.

MyMind, another app, is tempting as it offers a seamless web-clipper, auto-indexing, and has announced a graph view similar to Obsidian’s. Its auto-indexing feature uses AI to read through articles and even analyze pictures to identify and tag their contents. Essentially, it combines my Obsidian, Napkin, and Readwise workflows into one and manages to offer much more. The only issue with MyMind is the lack of metadata in its export feature, trapping users’ data within the platform. Moreover, it’s not as flexible as Obsidian. Taking complete ownership of data and privacy is essential to me, and I’m sure many of you were drawn to Obsidian for the same reason.

AI auto-tagging alleviates much of the friction in tracking our learning progress through Obsidian, making it invaluable. It allows us to focus on learning while the app takes care of the rest. Many platforms have already started taking initiatives, and I believe it’s time for Obsidian to embrace AI capabilities as well. If you agree, please leave a like for visibility.

For those concerned about AI integration compromising privacy in Obsidian, keep in mind that Obsidian is built upon strong principles. I’m confident that the Obsidian team can find a workaround to integrate AI without compromising privacy. A potential solution could involve an isolated repository of existing tags that the AI can reference and create similar tags or use existing ones. This way, AI will not have direct access to all the files in your vault, and users can manually summon the AI as needed.

PS: Just to clarify, this request is solely about auto-linking and not AI writing.

11 Likes
2

The primary problem is that this AI magic can’t be done locally with common hardware and with a cross-platform stack.

If you are willing to send your data to third-parties (OpenAI), there are already plenty of plugins.

I believe there are a handful of plugins (like graph analysis) that can assist you with link creation using classical machine learning (run locally)

2 Likes
3

I’m quoting this from a comment from my reddit thread

“What we can do is find the right balance by design, allowing users to control the data shared with third-party services. Imagine a scenario where I can choose what to share, like my web clippings that are processed by AI to tag, link, and much more, while my daily journal stays out of sight from AI. That way, I can focus on building my Knowledge Garden while the mundane task of organizing is taken care of by AI.”

I understand that running AI locally is challenging at the moment. I’m suggesting allowing third-party solutions within Obsidian by finding some middle ground when it comes to privacy concerns. I know Obsidian was founded on privacy-oriented principles. At the time of its launch, we hadn’t yet seen the full capabilities of NLPs. But now, it seems like the next big step, and Obsidian is a perfect fit for NLPs since it’s based entirely on markdown.

Being privacy-oriented doesn’t have to mean 100% offline. There’s a balance to strive for, and with the right precautions, we can maintain user privacy without depriving them of valuable functionality. I see no harm in sharing all the tags and selectively sharing notes.

As for the existing community plugins that work with OpenAI’s API, they do provide some functionality but are limited and don’t make much effort to preserve privacy. Users are left with no middle ground. The choices in front of them are extreme: either compromise on all the functionality or ignore privacy concerns completely.

If Obsidian decides to take steps towards integrating AI, the experience could be far more superior in terms of both functionality and privacy.

2 Likes
4

This wouldn’t require AI. This is natural language processing.
It has already been done for Joplin: GitHub - forcewake/joplin-tags-generator: Plugin for Joplin which can be used to extract keywords from note and assign them as a note's tags

The basic idea is to grab the words with the highest count.
Throw away words that don’t add meaning. (examples: is, a, the, or, etc)
Then give the remaining as choices or pick the top 5 or something.
It is kind of like what search engines do.

1 Like