lucasrosati/claude-code-memory-setup

What it solves

This project provides a comprehensive setup to give Claude Code persistent long-term memory and codebase awareness. It solves two primary issues: "amnesia" between sessions (where the user must re-explain the project) and high token consumption caused by Claude re-reading entire codebases to understand structure in every new session.

How it works

The system uses a multi-layered approach to manage context:

• Persistent Memory: An Obsidian Zettelkasten vault acts as a "second brain," storing decisions, architecture, and session logs. Claude Code interacts with this vault via a CLAUDE.md file and custom commands like /resume and /save.
• Codebase Mapping: It integrates Graphify to convert code into a queryable knowledge graph. Instead of reading raw files, Claude queries a structural map (AST-based), significantly reducing token usage.
• Chat Pipeline: A Python-based import pipeline exports Claude conversations (from both the CLI and Web app) into the Obsidian vault as indexed notes.
• Auto-Save Safety Net: A Python script registered as a SessionEnd hook in Claude Code automatically extracts session facts and writes a minimal log to the vault if the user forgets to manually save.

Who it’s for

Developers using Claude Code who want to reduce API costs (tokens) and maintain a continuous, searchable knowledge base of their project decisions and codebase structure across multiple sessions.

Highlights

• Token Efficiency: Claims up to 71.5x fewer tokens per session by replacing full file reads with graph queries.
• Zettelkasten Integration: Uses atomic, interlinked notes in Obsidian for declarative memory.
• Local Processing: Codebase graphs are processed locally via tree-sitter AST, ensuring code content doesn't leave the machine in default mode.
• Automated Sync: Includes cron-scheduled scripts to sync chat histories into the knowledge vault.