Skip to content

README.en.md

Latest commit

 

History

History
135 lines (93 loc) · 5.16 KB

README.en.md

File metadata and controls

135 lines (93 loc) · 5.16 KB
Raw

English | 日本語

EpisodicRAG Plugin

Hierarchical Memory & Digest Generation System (8 Layers, 100 Years)

EpisodicRAG Plugin - Architecture diagram of 8-layer hierarchical memory management system Version CI Tests codecov License

Overview

EpisodicRAG is a system that hierarchically digests conversation logs (Loop files) and structures them as long-term memory for inheritance. It automatically manages 8 layers of memory (Weekly → Centurial, approximately 108 years).

Key Features

  • Hierarchical Memory Management: Automatic digest generation across 8 layers (weekly to century)
  • Fragmented Memory Prevention: Instant detection of unprocessed Loops prevents memory gaps
  • Cross-Session Inheritance: Carry over long-term memory to next session via GitHub

Documentation Navigation

Your Goal Documents to Read
📚 Browse all documents INDEX.en.md
🚀 Get started QUICKSTART → Glossary
📘 Use daily GUIDE (Japanese)
📝 Quick reference CHEATSHEET
🔧 Customize settings digest-config (Japanese)
📊 Check status digest-auto (Japanese)
❓ Solve problems FAQ → TROUBLESHOOTING (Japanese)
🛠️ Contribute CONTRIBUTING → ARCHITECTURE (Japanese)
💡 Understand design philosophy CONCEPT
🤖 View AI/Claude specs AI Spec Hub (Japanese)
📋 Check changelog CHANGELOG (Japanese)

Note: Documents marked (Japanese) are available in Japanese only. Per our AI-First Documentation Policy, AI agents can understand and translate Japanese content on-the-fly.

Quick Installation

# 1. Add marketplace
/plugin marketplace add https://github.com/Bizuayeu/Plugins-Weave

# 2. Install plugin
/plugin install EpisodicRAG@plugins-weave

# 3. Initial setup (interactive)
@digest-setup

For detailed setup instructions, see QUICKSTART.en.md.

Basic Usage

Memory Retention Cycle

Add Loop → /digest → Add Loop → /digest → ...

By following this principle, AI can remember all Loops.

Main Commands

Command Description
/digest Detect and analyze new Loops
/digest weekly Finalize Weekly Digest
/dream-defrag Prune auto-memory (subtractive dream = GC; recommended when >50)
@digest-auto Check system status and recommended actions
@digest-setup Initial setup
@digest-config Change settings
@wakeup Session-start engine for claude.ai: loads long-term memory and applies the persona directive (requires config & Read token)

For details, see GUIDE.md (Japanese).

8-Layer Structure

Layer Time Scale
Weekly ~1 week
Monthly ~1 month
Quarterly ~3 months
Annual ~1 year
Triennial ~3 years
Decadal ~9 years
Multi-decadal ~27 years
Centurial ~108 years

For complete layer table, see Glossary

Cross-Session Memory Inheritance

With GitHub integration, you can retain and inherit long-term memory after session ends.

In claude.ai environments, the @wakeup skill automates session-start memory loading (SHA-pinned fetch via Read token; supports Private repo reference and PR write-back; all repo-specific values injected via config).

→ ADVANCED.md (Japanese)

License

MIT License - See LICENSE for details

Patent

Japanese Patent Application 2025-198943 - Hierarchical Memory & Digest Generation System

  • Personal/Non-commercial use: Freely available under MIT License
  • Commercial use: Please consult regarding patent rights before use

EpisodicRAG by Weave | GitHub