Membase

The memory API for AI agents

Give your agent a memory it keeps between runs, and share one memory between the agents you run. The same tools from MCP, the REST API and the Python and TypeScript SDKs: containers, memories and documents — search for passages, ask for an answer.

Unibase Memory

Unibase's Chrome client — encrypt, own, and verify cross-AI memory. Powered by Membase; identity via Unibase Pay.

Get started

Install the Membase SDK and give your agent a memory in a few lines. Membase can also be reached via MCP (Model Context Protocol) and skill for agent frameworks.

pip install membase-sdk

Quick start

from membase import Membase

client = Membase()          # MEMBASE_API_KEY from the environment

client.add("They want SSO before the pilot.", container="mv-...")
hits = client.search("what does Acme need before the pilot", limit=5)

Or clone locally: github.com/unibaseio/membase

How to use Membase

Step 1

Sync across platforms and devices

An open memory layer enables cross-platform, cross-device sync of conversations. Agents retain and build on prior interactions anywhere.

Step 2

Manage knowledge bases

Store and retrieve documents with embeddings. Add and query knowledge for RAG-style agent memory across platforms.

Step 3

Coordinate on-chain tasks

Architecture

Each account's memory lives in a runtime of its own rather than in a store shared across accounts, and one request exports all of it. Verifiable storage on Unibase DA is a pluggable backend in progress, not what runs today.

Resources

Membase

Unibase stack

Developers

FAQ

What is Membase?

Membase is a memory product for AI agents and the API behind it. People use it through the web app, a Chrome extension and a desktop app; developers reach the same memory through MCP, a REST API and the Python and TypeScript SDKs.

How do I use Membase from my own agent?

Install membase-sdk from PyPI or npm and create a client with a developer key, or connect any MCP-compatible client — Claude Code, Cursor and Codex included — to the hosted MCP endpoint. Both reach the same tools under the same permissions.

What is the Membase API, in one sentence?

Three nouns and two verbs: a container is a memory space, a memory is one fact, a document is material a container has read; search returns passages with the container each came from, and ask returns the agent's own answer.

Where does my memory actually live?

Each account's memory lives in a runtime of its own rather than in a shared store, and a single request exports all of it — the sources, the pages and the memory files — with a manifest stating what is included.