# Memorizz

A Python framework that connects persistent memory, model providers and reasoning so an AI agent can recall past context, reuse experience and coordinate with other agents across sessions.

Memorizz is a Python framework for building memory-first AI agents. It organizes memory into five kinds: episodic conversation history, semantic knowledge such as facts and preferences, procedural memory of registered tools and reusable workflows, short-term working context, and a shared workspace for coordinating between agents. Memory can be scoped by agent, user or conversation, stored locally on the filesystem, or backed by MongoDB, Oracle or Notion with a separate vector store.

Its MemAgent and MemAgentBuilder classes build persistent agents in assistant, workflow or deep-research modes, connecting to cloud or local model providers and to tools including Python functions, MCP servers, internet search, a governed browser and sandboxes; agents can delegate work to specialist agents and run scheduled automations. A separate MetaHarness component runs tasks through Codex, Claude Code, OpenHands or native Memorizz agents while carrying memory scope and learning evidence across whichever harness is used.

It is used through a Python SDK, an interactive command-line interface, a local web interface, or as an MCP server.

Website: https://github.com/RichmondAlake/memorizz