Second Brain Cc

Second Brain

Persistent memory and a working protocol for **any AI agent** — an [Obsidian](https://obsidian.md)-compatible vault of Markdown notes plus a small, standard-library-only Python engine: a SQLite FTS5 search index and a git-synced repo. Notes survive across sessions and machines, and a lightweight protocol tells the agent how to search that memory, isolate its work, and write back what it learned.

It is **agent-agnostic**. The knowledge and the engine are plain files and stdlib Python; agents connect through one of three thin integrations — an MCP server, a CLI, or a native Claude Code layer. Swap the agent, keep the brain.

Why

Agents start every session with a blank slate. They forget the decision you made last week, the convention this repo follows, and who the people and systems in your world are. This project gives an agent two things:

• Persistent memory — durable notes (decisions, conventions, how-tos, project state, people and systems) that outlive any single session.
• A working protocol — a repeatable loop: search the vault for context first, work in isolation, then save what changed.

Three ways to connect your agent

The vault is the same for all of them; pick whichever your assistant speaks.

Integration	For	Setup
MCP server	Any MCP agent — Claude Desktop, Cline, Cursor, Continue, Zed, Windsurf, your own client	Point it at integrations/mcp/server.py
CLI	Any agent that can run a shell, and you at a terminal	Put integrations/cli/brain on your PATH
Claude Code	The deepest experience: automatic recall, agents, slash commands	bash integrations/claude-code/install.sh

All three drive the **same** `_bin/` engine, so redaction, locking, indexing and the frontmatter contract hold no matter which agent is writing. You can use more than one, but avoid running two write-back integrations at once.