AI NewsProductivityAnnouncement
Second Brain OS is an Obsidian vault a Claude Code agent files and links on its own, and it has picked up 609 stars in 19 days
Second Brain OS is an MIT-licensed guide, vault template and Claude Code agent set that turns Obsidian into a knowledge base that files and links itself, and it has reached 609 stars and 101 forks in the 19 days since it went public.
Image: GitHub
Why it mattersThe bottleneck on a personal knowledge base has always been the filing work a human stops doing after two weeks, and handing that work to an agent decides whether a note system pays off or dies.
Anyone who has saved a hundred articles to read later already knows they will never open them again. The undefined-ui repository, Second Brain OS, argues that the filing and linking work is what kills a personal knowledge base, and hands that work to an agent so the vault stays alive.
The project went public on 7 September 2026 and has picked up 609 stars and 101 forks by 26 September, measured against the GitHub API, on an MIT licence. That is 19 days on the calendar. The repository is a full setup for one job: a starter Obsidian vault, an agent guide, and the skills and scripts an agent uses to keep the vault filed.
What is in the repository
The maintainer describes the shape in the README. There are 77 guide pages organised into ten sections that walk from the concept through setup, capture, structuring, retrieval, agents and maintenance. Alongside those, the repository ships 18 agent skills, 72 slash commands, and six subagents, four of which are read-only by design. There is a vault template with a wiki structure and a project pipeline, plus dependency-free Python scripts for link checking, vault stats and graph export.
The whole set is written for Claude Code and Obsidian, but the README states the design is portable. Every skill is a Markdown file the agent reads, every note is a Markdown file with links between notes, and Obsidian is a viewer for those links rather than a required tool.
The design decisions the README calls out
Three rules from the README are worth quoting because they show what the project is opinionated about.
Nothing is ingested until it is linked. A page that lands unconnected is invisible within a week, so linking happens in the same agent run as ingestion.
Contradictions are recorded, never overwritten. When a new source disagrees with a page, both positions stay, with dates and sources.
No fancy search index until you need one. The maintainer's view is that a personal notebook is small and already structured, so the wiki-style links answer most questions without extra tools running behind them.
The intellectual source is credited in the README: Andrej Karpathy's llm-wiki gist from 4 April 2026, plus two research papers on searching linked notes, and Sönke Ahrens on the Zettelkasten card method.
The consequence for a team is small and specific. Notes are a marketing team's raw material, a founder's log of what worked, and a working developer's memory across weeks. Those bases decay because filing takes daily effort nobody keeps up. An agent skill that files as it captures is a different bet than the last decade of note apps, and 32 stars a day suggest people want to make that bet. It runs on the machine, the files are portable Markdown, and walking away from every tool named in the repository leaves the folder intact.
Source
Repository: undefined-ui/second-brain-os on GitHub. Web version of the guide: undefined-ui.github.io/second-brain-os. Cited pattern: Karpathy's llm-wiki gist.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.