supermemory · @supermemoryai ↗

Supermemory

Give your AI lasting memory. One API adds fact extraction, live user profiles, and hybrid RAG search over your own data. Ranked #1 on LongMemEval, LoCoMo, and ConvoMem benchmarks.

28.3k stars Updated Jun 27, 2026 MIT 1 skill

The roster

What it does

Supermemory gives your AI persistent memory behind a single API, so it stops forgetting everything between conversations. It learns facts as you talk, keeps live user profiles that blend stable details with recent activity, and runs full RAG over your knowledge base, all from one call with no vector database or embedding pipeline to set up.

It handles the hard parts of memory: temporal changes, contradictions, and forgetting information that has expired. Connectors sync from Google Drive, Gmail, Notion, and GitHub, and multi-modal extractors read PDFs, images, and video. It ranks #1 on the LongMemEval, LoCoMo, and ConvoMem benchmarks, and you can call the API, use the app, or self-host it.

When to use it

  • Your AI forgets everything between sessions

    One API adds fact extraction, live user profiles, and search over your own data.

  • You need memory that retrieves the right thing

    Hybrid RAG search that ranks first on LongMemEval, LoCoMo, and ConvoMem.


Common questions

Can I self-host Supermemory, or is it a paid service?

You can do both. Supermemory is open-source under the MIT license and runs locally with npx supermemory local, and there is also a hosted API if you prefer not to run it yourself.

How do I add it to my AI?

You can connect Supermemory over MCP, install the SDK with npm install supermemory or pip install supermemory, or run it locally so the Memory API is available at localhost:6767.

Does my data leave my machine?

Not if you self-host. Running Supermemory locally keeps documents, memories, and search on your own machine, while the hosted option stores them in Supermemory's cloud behind an API key.

What does Supermemory actually do?

Supermemory extracts facts from conversations, builds live user profiles, and runs hybrid RAG search over your own data, so your AI keeps context instead of forgetting between sessions.