# Membase

> One memory. Every AI. Only yours.

Membase is a long-term memory for you and your AI. Bring knowledge, preferences and decisions from conversations, files and connected apps into a lasting memory you and your AI can build on. You choose what to keep and who can access it. Use the web app and Chrome extension today. Membase Desktop is coming soon, with local memory and offline search planned. Developers connect through MCP, REST, Python or TypeScript.

It is made by Unibase (https://unibase.com), which also publishes AIP, Unibase Pay and Unibase DA.

## Pages

- [Home](https://membase.com): what it is, how memory gets in, and who can read it
- [Changelog](https://membase.com/changelog): what shipped and when
- [Best AI Agent Memory Tools in 2026](https://membase.com/blog/best-ai-agent-memory-tools-2026): comparison of 9 agent memory tools by use case, license and price
- [How to Add Long-Term Memory to an AI Agent](https://membase.com/blog/how-to-add-long-term-memory-to-ai-agent): what to store, where it lives, and the SDK and MCP code to connect it
- [Mem0 Alternatives in 2026](https://membase.com/blog/mem0-alternatives): 8 AI memory tools compared with Mem0 by graph memory, self-hosting and price, with migration steps
- [Membase Use Case 01 · Coding Agents: Claude Code Persistent Memory](https://membase.com/blog/membase-use-case-coding-agents): make Claude Code, Codex CLI and Gemini CLI remember decisions across sessions with three hooks and one install command
- [Membase Use Cases 02: AI Agent Memory](https://membase.com/blog/ai-agent-memory-use-cases): three runnable Membase demos, persistent coding memory, shared memory for agent teams, and agents that learn from corrections
- [Best Memory APIs for AI Agents in 2026](https://membase.com/blog/best-memory-apis-for-ai-agents): 9 hosted memory layer APIs compared by free tier, pricing, billing unit and user isolation
- [Membase vs Mem0](https://membase.com/blog/membase-vs-mem0): which memory layer to use in 2026, compared by whose memory it is, access control, MCP, self-hosting, pricing and compliance
- [Extension privacy policy](https://membase.com/privacy): what the Chrome extension handles and where it lives
- [Documentation](https://docs.membase.com): guides and reference

## Use it

- Web app: https://app.membase.com
- MCP endpoint: https://api.app.membase.com/mcp-http
- Python: `pip install unibaseio-membase`
- TypeScript: `npm i @unibaseio/membase`
- CLI: `npx @unibaseio/membase-cli --help` — the binary is `membase`; add, search, ask, documents, profile
- Chrome extension: https://chromewebstore.google.com/detail/unibase-memory/edmncknbiihfoakimejbepnaeemaaamf

## Key facts

- **The model is three nouns.** Containers hold memories and documents. `search` returns passages with the container each came from; `ask` returns an answer over them. The same tool table is behind MCP, the REST API and both SDKs.
- **One memory, several agents.** Scoping is per container, so one agent can be given a project without being given everything else. Revoking access takes effect immediately.
- **It is yours to read and correct.** Review and correct saved knowledge, delete memories, export your data and revoke tool access.
- **Assistants it reaches through the extension:** ChatGPT, Claude, Gemini, Grok, DeepSeek, Kimi. Agents reach it through MCP or the SDKs.
- **Benchmarks** are for the memory engine, not the whole product: LoCoMo 93.1, LongMemEval 92.6, DMR 92.2, at roughly 6,500 mean context tokens against ~20,000 for full context.
- **Storage.** The hosted product keeps an account's memory in that account's own runtime. Unibase DA is available as a verifiable storage backend through the protocol SDK; the hosted product does not depend on it.
- **Data handling.** The hosted service stores and processes the sources you provide and is not end-to-end encrypted. See the extension privacy policy for the Chrome extension’s data handling.