agmemory 0.5.0

Persistent memory system for AI agents — session notes, rolling state, and RAG search

agmemory

Persistent memory system for AI agents — session notes, rolling state, and RAG search.

Features

• Session notes: Create structured markdown notes per session with metadata extraction
• Rolling state: Track focus, open items, decisions, pitfalls across sessions
• RAG search: BM25+ scoring with field-level weighting, query expansion, fuzzy matching
• MCP server: Expose all operations as MCP tools for AI agent integration
• CLI: Full-featured command-line interface

Installation

pip install agmemory

With optional features:

# Japanese tokenization support
pip install agmemory[japanese]

# Dense embedding search
pip install agmemory[dense]

# All extras
pip install agmemory[all]

Quick Start

CLI

# Initialize memory directory
memory init

# Create a session note
memory note new --title "Fix authentication bug"

# Search notes
memory search --query "authentication timeout"

# Show current state
memory state show

# Update state from a note
memory state from-note memory/2026-03-02/1830_fix-authentication-bug.md

MCP Server

Register in your .mcp.json:

{
  "mcpServers": {
    "memory": {
      "command": "memory",
      "args": ["serve"]
    }
  }
}

Or with uvx:

{
  "mcpServers": {
    "memory": {
      "command": "uvx",
      "args": ["agmemory", "serve"]
    }
  }
}

Memory Directory Structure

memory/
├── _state.md              # Rolling state
├── _index.jsonl           # Lightweight search index
├── _rag_config.json       # Search configuration
├── _vocab.json            # Vocabulary cache for fuzzy matching
└── YYYY-MM-DD/
    └── HHMM_slug.md       # Session note

License

Apache-2.0