One tweet with 500k views brought in 260 stars: how Second Brain OS turned a viral post into an "AI agent learning map" repo

Second Brain OS is a free markdown repo: an agent course with 7 modules, 5 handbooks, a glossary, 114 curated resources, a set of Claude Code tools, and a hands-on project to build a "second brain" that AI maintains for you, plus a clickable, rotatable knowledge-graph site. It was created in Sep 7 and has reached 6 stars and 1,449 forks by Nov 250, gaining 260 in the last two days. Almost all of those stars came from the audience the author built on X. Below: what it is, where the stars came from, why the same traffic produced so few stars, and how to think about building a similar project that can earn stars.

A tweet from Nov 3: "Lock yourself away for 15 minutes"

On Nov 3, 2026, X user Yarchi (@undefinedKi, 20K followers, self-described as "AI and tech researcher, data engineer") posted a tweet of just two sentences:

"Lock yourself away for just 15 minutes and go through this site. I promise your understanding of AI agents will exceed 95% of people."

Below it was a link, secondbrainos.dev. By 11-06, the tweet had 502k views, 11k likes, and 19k bookmarks (on X, 11-06).

The next day, AI news account @DataChaz, with 180k followers, wrote a long post saying the site "turns the whole field into a map you can browse, honestly impressive" (15k views). Two other accounts, @KanikaBK and @saidul_dev, also posted introductions. The repo's stars then jumped: 11-04 on one day, 1,189,11-05 +109,11-06 +151, the highest single-day total since the repo was created (per GitHub's daily star count).

Note that the tweet linked to the site, not the repo. 19k people bookmarked it, and roughly 260 people clicked through to GitHub and starred it over two days.

What it actually is

Open secondbrainos.dev and the homepage is a spinning sphere graph: 130 pages, 60k words, 57 terms, 107 tools, and 1,236 connecting lines. Each page, term, and tool is a bubble. Click one and the map rotates to it, lighting up everything it depends on.

The repo itself is plain markdown under the MIT license. You can read it without signing up or installing anything. The README opens by listing what's inside:

• Agent course: 7 modules, from "one prompt" to "a shipped agent," covering context, loops, when to skip expensive models, harnesses, evaluation, and deployment; 3 lessons per module, 21 pages in total;
• 5 handbooks: knowledge graphs, model routing, harnesses, loops, and evaluation, each going one level deeper, with a hands-on project at the end of each;
• Second-brain project: a beginner Obsidian vault with a built-in CLAUDE.md, so Claude Code can organize your notes, link them together, and answer questions about them;
• Toolkit: a Claude Code plugin with 18 skills, 72 slash commands, and 6 sub-agents;
• 114 curated resources: papers, repos, tools, and Obsidian plugins, with a monthly "new this month" section that says "verified at the source."

Its self-description has also changed. The repo's original tagline was "an AI second brain that maintains itself." The GitHub page now says "a guide and resource hub for AI engineers." The second brain has moved from the main attraction to one hands-on exercise inside the course.

How the stars built up: first, a long post with 9.16M views

The story starts with Jun 20, when Yarchi posted a long X article, "How to Build an AI Second Brain With Claude and Obsidian That Gets Smarter Every Day (Full Guide)." It walks you step by step through installing Claude Desktop and Obsidian, enabling plugins, and letting Claude read and write your notes vault. The opener says you can follow along even if you've never touched Claude Code or Obsidian. By 11-06 it had 9.16M views, 5,462 likes, and 20k+ bookmarks (X, 11-06). It builds on the "LLM Wiki" idea Karpathy floated in April, where an LLM maintains a markdown knowledge base for you. That was one of the hottest topics in the Claude Code crowd on X in 2026.

The timeline after that:

(View counts come from the X page as read on 11-06. Star counts come from GitHub, with one daily figure starting 10-17.)

Two conversion rates are worth noting:

• A long post with 9.16M views brought in 19 stars in the first 663 days after the repo was created, then slowly climbed to 1,100+ over the following month.
• A tweet with 500k views and 19k bookmarks brought in about 260 stars over two days.

People on X are willing to bookmark, but less willing to take the extra step to GitHub. The tweet led people to the site, and only a small "Repo" link in the top right corner of the site pointed to the repository.

It hasn't been on Hacker News or GitHub Trending (the repo is only two months old, and the daily 151 isn't enough to get there). The fork rate is high, though: 250 forks to 1,449 stars, about 17%, which is above what a typical bookmark-style repo gets. Looks like plenty of people are pulling the vault and running it themselves.

Who it's competing with

"Obsidian + Claude Code second brain" is one of the most crowded categories on GitHub this year, nearly all of it emerging after Karpathy's LLM Wiki:

Second Brain OS is a later entry in this wave. Its way of avoiding direct competition was to stop calling itself "yet another second-brain template" and present itself as an agent learning path with a map, something like an awesome-list that updates plus a tutorial series. That packaging shaped whether people were willing to bookmark it, but what got people to click in the first place was the article with 9.16M views and an account with 20k followers.

If it were you

One-line verdict: the "viral post → repo" path is real, and the "draw a field as a clickable map" format is fresh. But Second Brain OS's 1,449 stars came from an article with 9.16M views and a tweet with 500k views. Converted, that's not a high rate, and the second-brain category is already saturated. It's worth learning from, but not worth copying as-is.

Who it suits: people who already have a few thousand to tens of thousands of followers on X, Jike, Xiaohongshu, or Bilibili around a technical topic and can write long tutorials that get bookmarked widely; or people with real, systematic knowledge of a field (agents, RAG, evaluation, a specific framework) who can fill dozens of pages.

Who it doesn't suit: anyone planning yet another "Obsidian + Claude second-brain template." The top one has 16k stars, and the dozen behind it are stuck in the hundreds. It also doesn't suit people with no audience who expect the repo to be found through search. This repo never relied on search or Trending for much of its traffic.

How to do it:

• Change the topic, keep the format: don't build another second brain. Pick a field that doesn't have a "map" yet, such as a Chinese-language agent engineering path, the MCP ecosystem, or a panorama of a specific framework, and use the same structure: tiered courses, a few topic handbooks, a glossary, curated resources with a reason for each entry, and a clickable knowledge-graph site. (Inference: I didn't find a Chinese-language version of comparable completeness; not verified item by item.)
• Write the article first, then build the repo: June, Yarchi published the long post first and September created the repo later. Use a "follow along from zero" tutorial to test whether anyone bookmarks it, and turn it into a repo only once the bookmarks pile up.
• Copy its packaging details: the first screen of the README is a list of "what's inside" with numbers (how many modules, handbooks, and resources). A monthly "new this month" section gives people a reason to come back and share again. The spinning graph on the homepage is naturally suited to screen recordings for tweets.
• Fill the gap it left: its biggest tweet only linked to the site. Put a prominent "Star on GitHub" button on your site's first screen and at the bottom of every page, and give both the site and repo links in the tweet.

How to get users: on X, rely on your own long posts and short tweets, then have AI news accounts like @DataChaz (with hundreds of thousands of followers) repost once. For the Chinese version, post on Jike, Juejin, V2EX, and Xiaohongshu, and on Bilibili make a video titled "Tour this map in 15 minutes." The hook should be as short as theirs: "Spend 15 minutes and understand X better than 95% of people."

Window closes: the AI agent / Claude Code learning wave is still hot, and resource-roundup repos still get bookmarked now. It closes once the main platforms ship official learning paths, or once a dozen of these "full-map" repos show up at once and attention gets split. My estimate: a few months.

What would prove this wrong: ① a similar "learning map" repo without a big account pushing it hits a thousand stars on its own. That means the format itself earns stars and distribution matters less, which makes it more worth building. ② Second Brain OS keeps gaining dozens of stars a day from GitHub search and Trending instead of the author's posts. That means it has become the default entry point for the topic, and a same-topic repo adds little. You'd have to pick a different angle. ③ The author ships a Chinese or other-language version. Then the "build it in another language" path is gone.

Data and sources

GitHub repo page, README, and daily star/fork counts (10-17 through 11-06; the September numbers come from earlier points in the same record). secondbrainos.dev homepage (11-06). The author's long post on X (06-20) and tweets (10-07, 10-11, 11-02, 11-03), plus tweets from @DataChaz, @KanikaBK, and @saidul_dev (11-04). View, like, and bookmark counts were all read at 11-06. Competitor star counts are as of 11-06; the ones marked "~" are rounded.