Supermemory

Supermemory is a memory and context layer for AI agents and apps. It learns from conversations, extracts facts, builds user profiles, resolves knowledge updates and contradictions, forgets expired information, and returns the relevant context for each query. Document retrieval (RAG), connectors, and file processing sit in the same memory structure, so one system serves both knowledge-base search and personalized context.

Memory here differs from RAG: RAG retrieves the same document chunks for everyone, while memory tracks facts about a specific user over time and knows that a newer fact supersedes an older one. Supermemory runs both together by default in a hybrid search, and keeps an auto-maintained profile per user that splits stable facts from recent activity. Temporary facts expire once their date passes. Memory is scoped by projects (container tags), so work and personal context, or different clients and repositories, stay separate.

There are three ways in. For AI tools, plugins for Claude Code, Codex, OpenCode, OpenClaw, Hermes, Cursor, and Muse Code, plus a hosted MCP server, give an assistant persistent memory through memory, recall, and context tools. For builders, TypeScript and Python SDKs expose memory, profiles, search, connectors, and file upload through one API, with wrappers for AI SDK, LangChain, LangGraph, OpenAI Agents SDK, Mastra, Agno, and n8n. Supermemory local runs the same API as a single self-hosted binary with local embeddings and any model provider, including fully offline operation with Ollama.

On LongMemEval, Supermemory reports 95% Recall@15 while adding about 720 tokens of context, a 99.4% context reduction. MemoryBench, a companion open-source framework, runs standardized benchmarks of memory providers such as Supermemory, Mem0, and Zep.