Mnemosyne: memory tool for Hermes Agent

Hermes-first AI memory layer in one SQLite file with one pure-Python dependency and no cloud

What Mnemosyne does

Mnemosyne describes itself as a universal, Hermes-first memory layer. Installation is a single pip install, and memory lives in one SQLite database with one pure-Python dependency, so no cloud service is involved. For Hermes Agent it integrates through MCP and a plugin that the README says ships enabled. Other agents connect through MCP (Cursor, Claude Code, Codex CLI, Windsurf, or any MCP client over stdio or SSE), a native tool bridge for OpenWebUI, a Pi extension, an OpenClaw provider, or the Python SDK.

The Python API exposes remember and recall functions, and a CLI and an MCP server command are included. The README also has sections on architecture, a security and privacy model, a Hermes plugin and Mnemosyne Sync. It reports results on the BEAM benchmark but labels them as point-in-time measurements of v3.0.0 that have not been re-run, and says a LongMemEval figure was withdrawn until a reproducible rerun is published.

Key features

Single SQLite database with one pure-Python dependency and no external services
Native Hermes Agent plugin and MCP integration that ships enabled
MCP server for Cursor, Claude Code, Codex CLI, Windsurf and other MCP clients
Bridges for OpenWebUI, Pi and OpenClaw
Python SDK with remember and recall functions
Optional extras for vector search and the MCP server through the [all] install

When to use it

Give Hermes Agent persistent memory without running a cloud service
Share one memory store between Hermes Agent and Claude Code or Cursor through MCP
Add remember and recall calls to a custom Python agent

Who it is for: Developers who want local, self-contained agent memory that works with Hermes Agent first and other agent frameworks second.

How it fits with Hermes Agent

Built Hermes-first: Hermes Agent is integrated through MCP and a native plugin, and the repository is tagged hermes-agent and nousresearch.

How to install Mnemosyne

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

pip install mnemosyne-memory
pip install "mnemosyne-memory[all]"
pip install --upgrade mnemosyne-memory

FAQ

What is Mnemosyne?

Mnemosyne is a zero-cloud memory layer for AI agents. It keeps memory in one SQLite database, has one pure-Python dependency, and is described as Hermes-first while working with other agent frameworks.

Does Mnemosyne work with Hermes Agent?

Yes, the README lists Hermes Agent with MCP and a plugin, marked as native and shipping enabled. It also supports Cursor, Claude Code, Codex CLI, Windsurf, OpenWebUI, Pi and OpenClaw.

How do I install Mnemosyne?

Run pip install mnemosyne-memory. For all features, including vector search and the MCP server, use pip install "mnemosyne-memory[all]".