A Claude second brain is usually one of three setups: a Claude Project with uploaded knowledge and instructions, Claude's built-in memory, or Claude Code working inside an Obsidian vault of Markdown notes guided by a CLAUDE.md. Each works for one person inside Claude. We run the third one, twice over, and the part we've changed is where the shared half of it lives, so that our Grok Bots and everything else can read it too. Docs checked 2026-10-21.
What is the second brain in Claude?
Your notes, decisions and reference material kept where Claude reads them before it answers and, ideally, writes back to them. The test: can you open a new chat tomorrow, in any app, and have the agent know what you decided today? The concept comes from Tiago Forte's Building a Second Brain, "a way to cultivate a growing body of knowledge that is uniquely your own" (Building a Second Brain); what Claude adds is an agent that files, links and summarizes the notes for you.
Can a Claude Project or Claude's memory be your second brain?
A Project works as a reference library. It's a self-contained workspace with its own chats, knowledge base and instructions (Claude help: Projects); you upload documents and write instructions, Free accounts get five, paid plans switch large knowledge bases to retrieval, and each project has its own memory that doesn't read other projects' (Claude help: memory). Sharing one needs Team or Enterprise. The limit is that project knowledge is what you upload, and it lives in your Claude account.
Claude's memory helps, but it's Claude's summary of you, not your notes. It saves topics from your chats as you go, you can read, edit or delete each one, it's on by default on Free, Pro and Max, and it's shared between chat and Cowork when Cowork runs in the cloud. Claude decides what to save, so it suits preferences, not a decision log or meeting notes.
How do I set up a second brain in Obsidian with Claude Code?
Make a vault, run Claude Code inside it, and give it a CLAUDE.md that explains your system. Obsidian stores notes "as Markdown-formatted plain text files in a vault," a folder on your file system (Obsidian help), so Claude Code reads and edits them like any project.
- Create the vault in Obsidian (Manage Vaults..., then Create, or Open folder as vault for an existing folder).
- Open Claude Code in it.
cdinto the vault folder and runclaude. - Set the rules. Paste the prompt below, or write a short
CLAUDE.mdyourself, under 200 lines (Claude Code docs). - Feed it. Drop articles, transcripts and notes into
raw/and say "ingest the new files in raw/". - Read it in Obsidian, and commit it to Git for history and a second machine.
The pattern to borrow is Andrej Karpathy's LLM Wiki from April 2026: raw sources the LLM reads but never edits, a wiki of pages the LLM writes and maintains, and a schema file (CLAUDE.md for Claude Code, AGENTS.md for Codex) that sets structure and workflows, with an index.md listing every page and an append-only log.md. "Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase." The usual complaint about second brains is that they're hard to maintain, and this is why the setup holds up: the agent does the filing, and you only keep the rules short.
The prompt, run in a new vault folder:
Set up this folder as my second brain. Create raw/ for sources you never
edit, wiki/ for pages you write, index.md with one line per wiki page, and
log.md where you append one dated line per change. Put these rules in
CLAUDE.md, under 200 lines. When I add a file to raw/, summarize it into
wiki/, update index.md and log.md, and ask before deleting anything.
How ours works: two vaults, and a workspace for the shared part
We have two second brains, both Obsidian vaults with Claude Code running in them: a personal one and the company one, which holds the strategy docs, research, call notes and most of what Tokenrip has decided and why. Claude Code in the company vault is how a lot of our thinking gets done; the CLAUDE.md there is a short router pointing at the folders that matter for a given job, and the agent reads, writes and files like the wiki pattern above. We keep Claude Code's own auto memory off in both; Claude Code memory has why.
The change we made is about who else can read it. A vault on a laptop is invisible to Claude in the browser, to a Grok Bot on its own computer, and to a co-founder. So the parts of the vault that more than one agent or person needs are shared into a Tokenrip workspace, and a growing share of it has simply moved there; the plan is to move the whole vault over time. Our Grok Bots used to read the vaults directly and now mostly read the workspaces instead, and so does Claude on a phone, which is the other reason: a workspace is reachable on the go, and a folder on a laptop isn't. Obsidian stays the place for writing and reading; the workspace is the place the agents meet.
Why can't ChatGPT or Codex use my Claude second brain?
Because each setup lives inside your Claude account or on one laptop. Claude in the browser reaches outside data through connectors that connect from Anthropic's cloud, not your device (Claude help), so it can't open a local folder, and the same goes for ChatGPT. Codex can work in the vault, but it reads AGENTS.md, not CLAUDE.md, unless you configure a fallback (OpenAI); a symlinked AGENTS.md fixes that, and we do it. One user moving between Cowork and Claude chat summed up the rest: "When I shift surfaces during a workflow, context resets" (GitHub #55842).
| Setup | Who writes it | ChatGPT or Codex can read it? | Teammates | Main limit |
|---|---|---|---|---|
| Claude Project | You upload, Claude reads | No | Team and Enterprise only | Grows only when you upload |
| Claude memory | Claude, with your edits | No; experimental import and export | No | Claude chooses what to keep |
| Claude Code and an Obsidian vault | You and Claude Code | Codex, in the vault with an AGENTS.md |
Through Git or a sync service | Web apps and other machines can't see a local folder |
| Notion with Notion MCP | You and any MCP app | Yes, with your full Notion permissions | People with access | Notion says it works best for individuals |
| A shared workspace | Every agent you connect | Yes | Yes, with their own agents | One more place to connect |
For one person mostly in coding agents, Git plus a shared AGENTS.md goes a long way. It doesn't reach ChatGPT, Claude on your phone or a teammate's agent without manual uploads, and nothing tells the next agent what changed.
Sharing the second brain with every agent
The split we use: Obsidian for your own writing, a Tokenrip workspace for what every agent must share, meaning decisions, handoffs, tasks and the files behind them, versioned with diffs. Claude on web, desktop or mobile connects through Customize > Connectors > Add custom connector with https://api.tokenrip.com/mcp and the emailed code, Claude Code with claude mcp add --transport http tokenrip https://api.tokenrip.com/mcp, and ChatGPT or Codex through their own routes. Pin the operating rules and current decisions, up to 20 markdown documents every agent reads first. Or let any agent with a shell do the setup with this line; you give an email address and a six-digit code.
Connect me to Tokenrip. Fetch https://tokenrip.com/setup?ref=claude-second-brain and follow it.