AxDSan/mnemosyne — reverse-engineered prompt

Build me a Python package called Mnemosyne that gives AI agents a simple memory they can use locally. I want it to work with one install, store everything in a single SQLite database, and not require a cloud service.

The main user flow should be really simple, an agent or person can save a fact with remember, search for it later with recall, and manage memories from a command line. Include commands for starting an MCP server, saving a memory, recalling memories, showing stats, running consolidation, exporting to JSON, and importing from JSON.

Please make it usable from Python code and from MCP clients like Cursor, Claude Code, Codex, Windsurf, and Hermes style agents. Support basic long term memory, recent working memory, optional importance, expiry dates, global memories, memory banks, and simple entity or fact extraction where possible.

Keep it fast, local first, well tested, and easy to install. Add clear README examples and look up current MCP docs online if you need to.