Second Brain

Second Brain is an open-source, self-hosted memory layer built by rahilp that gives AI tools like Claude, ChatGPT, and Cursor a shared continuity layer across sessions. It deploys to your own Cloudflare account in minutes using a one-click button, and stores context — decisions, project notes, preferences — that can be recalled semantically across every MCP-compatible client. The project is licensed under MIT and runs entirely on Cloudflare's free tier at personal scale.

What It Is

Second Brain solves the problem of AI tools starting from zero every session. Instead of re-pasting context into each new conversation, users store information once and recall it by meaning across any tool that supports the Model Context Protocol (MCP). The project positions itself as a continuity layer rather than a simple note-taking app: memories are searchable by semantic meaning, time-decay ranked so recent decisions surface first, and fully owned by the user in their own Cloudflare account.

How the Memory Layer Works

The core MCP server exposes six tools: remember, append, update, recall, list_recent, and forget. The recall tool uses semantic search powered by Cloudflare Vectorize and the bge-small-en-v1.5 embedding model, so asking "what did I decide about the pricing model?" surfaces the right note even without exact keyword matches. Long notes are split into overlapping chunks so specific paragraphs can surface precisely. Near-duplicate detection blocks redundant entries before they are stored.

Ecosystem and Capture Points

The project ships with multiple capture integrations so memory gets filled naturally:

• CLI — npm install -g second-brain-cf-cli for brain remember and brain recall from the terminal
• Obsidian plugin — available in Obsidian Community Plugins as second-brain-sync; supports auto-sync on save, sidebar search, tag filtering, and selective sync
• iOS Shortcuts — Brain Dump (voice), Text Brain Dump, and Save to Brain shortcuts
• Browser extension — captures any page or highlighted text from Chrome
• Bookmarklet — lightweight alternative for any browser
• Web dashboard — browse, search, and edit memories at the Worker URL; works from mobile browsers

Architecture and Stack

The entire stack runs on Cloudflare primitives: Workers for compute, D1 SQLite for structured storage, Vectorize for the vector index, and Workers AI for embeddings. The MCP server is implemented in TypeScript using the MCP TypeScript SDK. Schema auto-creates on first request, and no credit card is required for personal-scale use. All data stays in the user's own Cloudflare account — no third party touches stored memories.

Update: v1.6

The project reached v1.6 (published 2026-05-31), with the repository last pushed on the same date. The project page reports v1.5.1 on the marketing site while the GitHub release tag is v1.6, indicating active iteration. The repository had 110 stars and 24 forks at the time of the latest data snapshot, and the project launched on Product Hunt during this period. Development activity and open issues (10) suggest the project is in active early-release development.