Original video in Chinese.
Key Takeaway
- Major Obsidian CLI update: directly operate Obsidian from the terminal command line, achieving almost all in-app functionality. It is designed for AI, letting AI invoke a note repository as efficiently as a human.
- Solves two core pain points: search is upgraded from “blind keyword traversal” to Obsidian’s native indexing (supports regular expressions, tags, frontmatter); it fully perceives note relationships (backlinks, knowledge graph), turning a “pile of files” into an intelligent knowledge base.
- The best solution in the AI era: I have already integrated it into newtype CLI and plan to support it at the underlying level; Obsidian firmly holds the top spot among note-taking apps, and other tools that do not support local storage plus AI invocation will be eliminated.
Recently, Obsidian launched a new feature: CLI.
I think this is Obsidian’s most important update in the AI era.
Let me talk about this feature first. Put simply, you can directly operate Obsidian from the terminal through command line commands and achieve the vast majority of the app’s functions.
For example, if I enter obsidian create, it creates a note with both title and content empty. If I enter obsidian create name="test" content="123", it creates a note titled test with content 123.
That’s just the simplest example. Through the command line, almost everything you can do in the app can also be done in the terminal.
So the question is: why not just use the app directly and insist on typing commands? Is this the Renaissance? Is it just putting on airs?
The answer is simple:
The app is for humans to use, while commands are for AI to use.
In fact, before this, AI had already started taking over note repositories. Because Obsidian has always followed a Local First philosophy, and every note is a Markdown document stored locally, from AI’s perspective a note repository is just an ordinary folder, nothing special. If AI can edit a Python script, then naturally it can also edit a note document.
I should be one of the first bloggers in China to advocate for using AI to take over Obsidian. At first, I used AI plugins. About more than 10 months ago, I proposed using Cursor to open the note repository to enable note retrieval and editing. About 8 months ago, I proposed using Claude Code to build a personal AI operating system.
Over the past three months, Skills became popular in the Chinese-speaking community and also made Claude Code popular. As a result, more and more people started using it to build an AI OS.
The starting point for using Cursor or Claude Code is exactly what I just mentioned: letting AI coding tools directly operate the file system. There is nothing wrong with this approach, but it is still not good enough. I think there are mainly two drawbacks.
First, the search capability is limited.
If AI directly operates the file system, it can only find file paths through keyword matching and then read them.
Now that there is a command line tool, AI can directly invoke Obsidian’s native search engine. This engine is much more powerful than simple keyword search, because it supports regular expressions, tag filtering, frontmatter property queries, and more.
Also, from a speed perspective, Obsidian has already built an index for the entire repository, so searching through it directly will definitely be faster than traversing files.
Second, it has no awareness at all of the relationships between notes.
If AI directly operates the file system, what it sees are only individual Markdown documents.
But Obsidian notes can be connected to one another. For example, through the well-known backlink feature, one note can reference another. Relationships like this cannot be discovered by directly operating the file system.
Now that there is a command line tool, AI can use simple commands to know which notes a note references and which notes reference it.
So for AI, what it faces is no longer just a pile of documents, but a knowledge graph. The note relationships you built in Obsidian before can all become useful in the AI era.
These two points I just mentioned are, in my view, the qualitative leap brought by Obsidian CLI. It changes how AI coding tools use a note repository from the old “blindly traversing the file system” approach to “operating a knowledge base with indexing and a graph of relationships.”
This is extremely, extremely valuable.
So I’ve already configured it in newtype CLI.
First I gave Grok the official documentation link and asked it to compile a list of commands. Then I pasted both the link and the list into newtype CLI, and after Chief digested them, I had them saved into the AGENTS.md document. That way, the next time it needs to use the command line to operate Obsidian, it will have the correct reference.
When the next newtype CLI update comes out, I plan to add these commands as well and implement support at the tool’s underlying level.
Now that this newly released CLI feature is here, Obsidian can be said to firmly hold the top spot among note-taking apps in the AI era. I have one intuition: any note-taking app that does not support local storage and does not support invocation by other AI tools will not survive in this AI era. Let’s wait and see.
OK, that’s all for this episode. If you want to learn about AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!
Keep Practicing
With a tool interface in place, you still need to be clear about how materials enter the system and how they are used. You can compare different access methods with the Obsidian AI workflow, and then use the Guide to Building a Wiki for Creators to organize source materials, page relationships, and long-term maintenance rules.