ponymemory

<p align="center"> <h1 align="center">🧠 PonyMemory</h1> <p align="center"> <strong>Persistent, cross-session memory for Claude Code</strong><br> Semantic search + Knowledge graph — fully local, zero API keys </p> <p align="center"> <a href="https://ponytech-dev.github.io/ponymemory/">🧠 Live 3D Brain Atlas</a> • <a href="#quick-start">Quick Start</a> • <a href="#how-it-works">How It Works</a> • <a href="#mcp-tools">MCP Tools</a> • <a href="#architecture">Architecture</a> </p> </p>

---

> [🧠 Interactive 3D Brain Atlas →](https://ponytech-dev.github.io/ponymemory/) Explore the 5-layer memory architecture as a 3D brain — each region maps to a real neuroscience structure.

---

Claude Code forgets everything between sessions. Every new conversation starts from scratch — no memory of your preferences, past decisions, project context, or the corrections you've already given.

PonyMemory makes Claude Code remember. It automatically stores facts, decisions, and corrections during conversations and retrieves them in future sessions. No manual tagging, no copy-paste, no maintenance.

Session 1: "Don't mock the database in tests, use real connections"
    → PonyMemory stores this as a correction

Session 47: (You start writing tests)
    → PonyMemory auto-injects the correction
    → Claude Code writes tests with real DB connections

Unlike systems that add a separate LLM layer for memory management, PonyMemory lets Claude Code itself decide what to remember, detect duplicates, and resolve conflicts. This means:

• No extra LLM costs — Claude Code handles the intelligence
• No separate memory service — just storage backends (Qdrant + Neo4j)
• Better judgment — Claude Code understands context better than a generic extraction pipeline

• Semantic Memory — Vector search via Qdrant finds relevant memories by meaning,