Second Brain for AI deserves credit up front: it is an MIT-licensed project that made the same argument we make (memory for your AI tools should be under your control, not locked inside one platform) and it shipped that argument months before we launched. Both tools give Claude, Cursor, and other MCP clients persistent cross-tool memory under self-custody. The interesting differences are in where the memory lives, what shape it takes, and what the tool is really for.
At a glance
| Second Brain for AI | Bourdon | |
|---|---|---|
| Where memory lives | Your own Cloudflare account (Workers, D1, Vectorize) | Plain YAML on your own disk |
| Memory shape | Semantic store: embeddings, vector search, memory graph | Structured library in a published schema: projects, sessions, entities |
| Retrieval model | Semantic search: recall by meaning | Recognition-first timing: ms-scale recall, background hydration (L0-L6) |
| Center of gravity | Personal knowledge capture: Notion, calendar, email, browser, Obsidian | Agent runtime memory: 16 shipped participants over MCP |
| Interop contract | Its own API and store | Apache-2.0 wire format + schema anyone can implement |
| Federation | Single deployment | Optional peer federation, strictly opt-in, quarantined receive path |
| License | MIT | BUSL-1.1 engine (converts to Apache-2.0 per version), Apache-2.0 interop |
| Price | Free, self-hosted | Free, self-host forever |
Second Brain for AI claims are drawn from its public repository README and Product Hunt listing as of 2026-08-04. If we’ve mischaracterized anything, corrections welcome at [email protected].
Cloud you rent vs disk you sit at
Both projects reject the hosted-account model, which puts them on the same side of the category’s biggest divide. They land in different places within self-custody: Second Brain runs as a service on your Cloudflare account: real ownership, reached over the network, stored as embeddings and rows. Bourdon’s library is files on the machine where your agents run: readable without Bourdon installed, diffable under version control, and portable by cp.
That difference matters most at the exits. Leaving a semantic store means exporting and re-embedding somewhere else. Leaving Bourdon means keeping the directory you already have.
Different questions, honestly
Second Brain is strongest as a capture tool: it pulls from Notion, calendars, email, and the browser, then lets any AI tool recall it by meaning. Bourdon is strongest as a runtime layer: agents writing and recognizing shared working context with millisecond timing while they operate. A fair reading is that they answer different questions that happen to share a category page.
When Second Brain for AI is the better fit
- You want semantic search over free-form knowledge (notes, emails, meetings) from any AI tool.
- You’re already on Cloudflare and like Workers + D1 + Vectorize as a personal stack.
- MIT licensing end to end is a requirement.
- Capture breadth (Notion, calendar, email, browser extension, Obsidian) matters more than agent runtime behavior.
When Bourdon is the better fit
- You run multiple coding agents and want them sharing one structured library with a published contract, not a private store.
- You want memory as literal files: greppable, git-able, readable without the tool.
- Response timing matters: recognition in milliseconds, deeper context hydrating in the background.
- You may eventually want multiple machines or peers federating, with quarantine and validation on everything inbound.
Try it
pip install bourdon && bourdon setup
# or the MCP server via npm:
npx @getbourdon/mcp-server --library ~/agent-libraryStart at the landing page, or read the recognition-first thesis for the full runtime argument.