Getting Started¶
Memorg is distributed as the memorg package on PyPI. It bundles a Python library, an interactive CLI (memorg), and an MCP server (memorg-mcp).
Prerequisites¶
- Python 3.11 or 3.12 — declared in
pyproject.toml - OpenAI API key — used for embeddings (
text-embedding-ada-002, 1536-dim) and chat completions
Choose Your Interface¶
Memorg can be used in three different ways. Pick the one that matches your workflow:
| Interface | When to use it | Entry point |
|---|---|---|
CLI (memorg) |
Quick experimentation, single-user chat-with-memory | memorg.cli_entry:main |
| Python library | Embed memory in your own application or agent loop | from memorg import MemorgSystem |
MCP server (memorg-mcp) |
Expose memory to Claude Desktop or other MCP clients | memorg.mcp.cli:main |
All three interfaces sit on top of the same MemorgSystem core, so a database written by the CLI can be read by the library or the MCP server, and vice versa.
Storage Footprint¶
A fresh Memorg install creates two artifacts on disk (default base name memorg):
memorg.db— SQLite database holding sessions, conversations, topics, exchanges, and FTS5 indexesmemorg.usearch— USearch vector index (1536-dimensional, cosine metric)
Both files share the same base path, so passing --db-path ./data/memorg.db writes ./data/memorg.usearch next to it.
Next Steps¶
- Installation — install from PyPI or source
- Quick Start — first session, conversation, topic, and search
- Configuration — environment variables and component options
Support¶
- Issues: github.com/neul-labs/memorg/issues
- Architecture: Overview, Technical Specification