The Obsidian CEO's repo should be ranked ahead of 11 AI memory projects

Ryven compiled a list on X: 11 repos, with only one goal: letting AI agents truly read and write your Obsidian vault. The original post also included a video that called this SECOND BRAIN, with the 11 repositories corresponding to stages such as capture, notes, and connections.

The one most worth looking at first in the list is 01 obsidian-skills. It comes from the GitHub account of the Obsidian CEO, kepano. Its role is very specific: teaching Claude Code, Codex, and other agents to write notes, bases, and canvases in the way Obsidian expects.

This point is often overlooked. When many people talk about AI memory, their first reaction is to have the model remember more chat content. But when an agent works in your vault, it first needs to understand how files are organized, how properties are written, and what structure bases and canvases have. If it can't write into the vault, neither memory nor retrieval can connect.

So the more stable order is: first teach the agent to write, then get it running, and only after that talk about what it should remember.

02 obsidian-copilot solves the getting-it-running problem. It brings OpenCode, Claude Code, and Codex into Obsidian, supports bringing your own key, and also supports local models. The official description recommends OpenCode; it can be downloaded and managed, and it can also connect to your own provider.

03 supermemory provides the memory layer. It extracts facts from conversations, maintains user profiles, handles knowledge updates and contradictions, and automatically forgets outdated information. The README gives several metrics: it ranks first on the three memory benchmarks LongMemEval, LoCoMo, and ConvoMem, with 95% Recall@15, 99.4% context reduction, and user profiles at about 50ms. It can also run locally as a single binary, with data stored in ~/.supermemory.

04 khoj leans toward Q&A. You can use your own server to ask questions of notes, PDFs, and webpages.

05 obsidian-smart-connections finds notes you forgot you had written. It automatically links them as you type, with no API key required.

06 karakeep handles saving. Throw webpages, content, and files into it; AI tags them, and later you retrieve them on your own server.

07 siyuan lets users and AI agents edit the same set of notes, and it can also be self-hosted.

08 zk is a plain-text wiki in the terminal. 09 silverbullet uses Lua to add programming capabilities to Markdown notes. 10 memos is timeline-style quick capture, retrieved via search, tags, and dates. 11 trilium supports a personal knowledge base with a hierarchical structure.

The list is comprehensive, but two reminders in the replies are more useful than the list itself.

A user named Peter said he would handle old-memory import first. He migrated 122 Claude Code memory files into an Obsidian vault and verified them byte by byte: 240438 bytes in, 240438 bytes out. He later changed 25 filenames from underscores to hyphens and fixed 21 broken links. This detail shows that migrating old memory is dirty work that has to be done before installing plugins.

Another user, Ruslan, mentioned contradictory memories. When a new note overturns an old note, he wants both sources and dates preserved. 'Memory updated' should not quietly turn unresolved disagreements into facts. Supermemory's description does say it automatically handles knowledge updates and contradictions, but when it comes to a personal knowledge base, preserving sources and dates is still worth insisting on.

A user named Road N put it more directly: that official obsidian-skills is the safe choice, while memory-type plugins tend to get messy on large vaults. An Jiaoshou's comment was that Obsidian is almost being turned into an agent's second brain.

The value of this list is not in its quantity. It pushes Obsidian one step further from a note-taking tool toward a file layer that agents can operate on. If you want to try it, install obsidian-skills first, or use obsidian-copilot to get an agent running first. Try memory-type projects on a small vault first; don't connect them all at once right away. Migrating old memories, sourcing contradictions, and fixing broken links—these unglamorous-looking tasks are what determine whether a second brain can be used long term.

Original post: https://x.com/imryven/status/2117017504321282498

Repo list:

01 obsidian-skills https://github.com/kepano/obsidian-skills

02 obsidian-copilot https://github.com/logancyang/obsidian-copilot

03 supermemory https://github.com/supermemoryai/supermemory

04 khoj https://github.com/khoj-ai/khoj

05 obsidian-smart-connections https://github.com/brianpetro/obsidian-smart-connections

06 karakeep https://github.com/karakeep-app/karakeep

07 siyuan https://github.com/siyuan-note/siyuan

08 zk https://github.com/zk-org/zk

09 silverbullet https://github.com/silverbulletmd/silverbullet

10 memos https://github.com/usememos/memos

11 trilium https://github.com/TriliumNext/Trilium