gitmyhub

mnemo-agent

Python ★ 0 updated 9d ago

🗄️ Mnemo — an AI assistant with a persistent, temporal knowledge-graph memory (Graphiti + FalkorDB). Remembers across Claude Code sessions via MCP tools + hooks; recalls who/what/when with citations.

An AI memory tool that lets Claude Code remember facts, people, and dates across separate sessions using a graph database.

PythonFalkorDBGraphitiDockerMCPsetup: moderatecomplexity 4/5

Mnemo is an AI assistant built to remember things across separate Claude Code sessions. Normally, when you close a chat with an AI assistant, everything you told it disappears. Mnemo fixes that by storing facts in a memory database that persists on disk, so you can tell it something today and ask about it again next week, in a brand new conversation, and it will recall the answer along with who said it and when it was true.

Under the hood, Mnemo pairs a graph database called FalkorDB with a tool called Graphiti that turns plain text into a network of connected facts: people, projects, and the relationships between them. When you tell Mnemo something, it extracts the entities and links them together with dates attached. When you ask a question, it searches that graph using a mix of keyword matching, vector similarity, and relationship traversal, then answers in the voice of an archivist, always citing where and when a fact came from.

Mnemo connects to Claude Code through the Model Context Protocol, exposing simple remember and recall tools, plus hooks that automatically start the memory database when a session begins and summarize the conversation into memory when it ends. This means a developer does not have to manually save notes. The embeddings used for search run locally on your machine, so no separate API key is needed for that part, though Claude itself is still used to extract facts and generate answers.

To try it, you clone the repository, add your Anthropic API key to a config file, start the memory database with Docker, install the Python dependencies, and run the included test suite to confirm everything works. The project is explicitly described as an experiment rather than a finished product: it supports only a single local user, has no hosted or multi-user version, and its automatic session summaries can be uneven in quality. It is aimed at developers who already use Claude Code and want their coding assistant to remember context between sessions instead of starting from zero every time.

Where it fits