Updates·November 2, 2026, 10:31

One shared memory aims to follow you across AI tools

AI-generated from the sources listed, with automated source checks.

The Membase project wants to give users a single store of memory that multiple AI tools, such as Claude, ChatGPT and Cursor, can read from. The features are the developer's own descriptions.

If you switch between several AI tools, you know the drill: You have to explain the same things over and over. The Membase project wants to fix that with a single shared memory layer that all your AI tools can use.

According to the project itself, the idea is that you write a memory once, and all connected tools then read from the same store. The slogan is: one memory for all AIs.

How it is meant to work

Users can add facts, documents, web pages and exported chats to the shared memory. It can be divided into containers, for example one per project, and users can search it, ask it questions and manage it.

According to the project's own site, Claude, Claude Code, ChatGPT, Cursor and Codex can access the same memory. This happens either through SDKs (ready-made software packages for developers) or through MCP connections. MCP, the Model Context Protocol, is a common standard that lets AI tools connect to external data sources.

Integrations are also listed for agent frameworks such as LangChain, Microsoft Agent Framework, Dify, Hermes Agent, OpenClaw and Strands Agents. These are toolkits that developers use to build AI agents.

Hosted or local

Membase can run in hosted mode, which requires an API key, or in local mode. SDKs are available for Python (version 3.10 and later) and TypeScript (Node 18 and later), along with a command-line interface.

A comparison from competitor EverMind describes Membase as a managed personal memory layer in the Unibase ecosystem. Here, knowledge is structured as a graph, where personal memory and knowledge are kept separate and organized into episodes and wiki documents. According to the same description, data is hosted on Unibase's infrastructure and not on the user's own servers.

Caveats

The features come from Membase's own site and are therefore the developer's own claims. Membase itself reports 93.1% on LoCoMo, 92.6% on LongMemEval_S and 92.2% on DMR. These are its own figures, which have not been independently verified. EverMind, which publishes a competing solution itself, claims that there are no comparable peer-reviewed or public results. HaluMem is not mentioned among Membase's own figures. These are tests that measure how well AI memory works.

The GitHub page lists no launch date.