Mnemosyne is a Python-based memory system that uses a single SQLite database and a pure-Python dependency. It supports MCP and Python SDK usage with a BEAM architecture and optional client-side encryption for sync.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Mnemosyne is a universal memory layer designed to work with Hermes-first integration across multiple agent frameworks. It uses SQLite as the storage backend, aims to run with zero external services, and offers a Python SDK and MCP-based server for memory operations.
How it works
The architecture centers on BEAM (Working Memory, Episodic Memory, TripleStore) layered over a SQLite database. It provides direct SQL-backed storage with vector search capabilities via sqlite-vec and FTS5, and supports a hybrid scoring scheme combining vector similarity, FTS ranking, and importance.
Getting started
Conscious of the README commands:
pip install mnemosyne-memory
# With all features (vector search + MCP server)
pip install "mnemosyne-memory[all]"
# Upgrade
pip install --upgrade mnemosyne-memory
Recent releases
Latest release: v3.15.1 (2026-07-30) with changes pinning remaining GitHub Action refs and closing persona adapter transactions. Earlier release: v0.5.0 (2026-07-19) addressing a security write-approval bypass and path traversal. Other listed items include v3.14.0 (2026-07-17) and v3.12.2 (2026-07-11).
Traction
Stars: 2069; Forks: 173; Open issues: 68.
Behind the repo
No company-level partner details are included beyond the sponsor note in the README.
Caveats
License: MIT. Created: 2026-04-05. Last push: 2026-08-04. Language: Python. Environment variables and configuration options are documented under Environment Variables in the Configuration section.