supermemoryai/supermemory

What it solves

Supermemory addresses the "forgetfulness" of AI assistants by providing a persistent memory and context layer. It prevents AI from losing information between conversations, allowing it to remember user preferences, project details, and past discussions without requiring the user to repeat themselves.

How it works

Supermemory acts as an engine that automatically extracts facts from conversations, builds and maintains user profiles, and manages knowledge updates. It distinguishes between RAG (retrieving static document chunks) and Memory (tracking evolving facts about a user). The system handles temporal changes, resolves contradictions, and automatically forgets expired information. It can be integrated via a single API, an MCP server for AI tools, or run as a local binary with support for various LLM providers and local embeddings.

Who it’s for

• AI Tool Users: People using assistants like Claude Code, Cursor, or VS Code who want their AI to have persistent memory across sessions.
• AI Developers: Engineers building agents or apps who need a turnkey solution for memory, RAG, user profiles, and data connectors without managing vector databases or embedding pipelines.
• Self-hosters: Users who want to run a state-of-the-art memory engine locally on their own machine for privacy or offline use.

Highlights

• Top Benchmark Performance: Ranked #1 on LongMemEval, LoCoMo, and ConvoMem benchmarks.
• User Profiles: Automatically maintains static (long-term facts) and dynamic (recent activity) profiles for fast context injection (~50ms).
• Hybrid Search: Combines RAG and personalized memory in a single query.
• Broad Connectivity: Built-in connectors for Google Drive, Gmail, Notion, OneDrive, and GitHub.
• Multi-modal Extraction: Processes PDFs, images (OCR), videos (transcription), and code (AST-aware chunking).
• Flexible Deployment: Available as a cloud API, MCP server, or a zero-config local binary.