Artemis Memory System

A-Mem inspired memory system with MCP integration for AI agents. Provides dynamic memory evolution, context engineering, and native Claude Code integration.

Features

• A-Mem Style Notes: Atomic, interconnected memories with auto-linking
• Dynamic Evolution: New memories automatically link to and evolve existing ones
• Context Assembly: Task-specific context injection with token budgets
• MCP Integration: Native Claude Code tools for memory operations
• Session Briefings: Generated summaries of projects, activity, and insights

Installation

cd ~/Projects/artemis-memory
pip install -e .

Claude Code MCP Configuration

Add to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "artemis-memory": {
      "command": "python",
      "args": ["-m", "artemis_v2.mcp_server"],
      "cwd": "/Users/loganbeharry/Projects/artemis-memory"
    }
  }
}

Or for Claude Code CLI, add to ~/.claude.json:

{
  "mcpServers": {
    "artemis-memory": {
      "command": "python",
      "args": ["-m", "artemis_v2.mcp_server"],
      "cwd": "/Users/loganbeharry/Projects/artemis-memory"
    }
  }
}

MCP Tools

Tool	Description
memory_add	Store a new memory with tags and context
memory_search	Search memories by keyword or tag
memory_recall	Quick topic-based recall
context_for_task	Get assembled context for task types
project_status	Get/update project context
session_briefing	Generate session briefings
memory_evolve	Trigger memory consolidation
preference_set	Set user preferences

Usage Examples

Adding Memories

memory_add(
  content="Prefer functional patterns over OOP in Python",
  tags=["preference", "python", "coding"],
  context="Observed during code review"
)

Searching

memory_search(query="python patterns", tag="preference")

Getting Context

context_for_task(task_type="coding", project="my-project")

Architecture

artemis_v2/
├── memory/
│   ├── note_manager.py    # A-Mem note operations
│   └── evolution_engine.py # Linking and evolution
├── context/
│   └── assembler.py       # Context engineering
├── briefings/
│   └── generator.py       # Session briefings
└── mcp_server.py          # MCP integration

Data Storage

Memory data is stored in ~/Projects/artemis-memory/data/ (configurable via ARTEMIS_MEMORY_DIR env var):

data/
├── memory/
│   ├── notes/             # Individual note JSON files
│   ├── index.json         # Search index and link graph
│   └── evolution_log.json # Evolution history
├── context/
│   ├── projects/          # Project context files
│   ├── skills/            # Skill/knowledge files
│   └── preferences/       # User preferences
└── briefings/
    └── current_context.md # Latest session briefing

Running Tests

pip install -e ".[dev]"
pytest tests/

Research References

• A-Mem (NeurIPS 2025) - Zettelkasten-inspired memory architecture
• Letta Benchmark - Filesystem approach validation
• Mem0 - Production memory patterns