What is superdupermemory?
AI agents are stateless by default. Every conversation starts from scratch. superdupermemory gives your agents a persistent memory — facts extracted from conversations, stored locally, recalled with semantic search.
How it works
- You send text — a conversation turn, a user message, anything
- Facts are extracted — an LLM reads the text and pulls out discrete facts (
user.name: Alice,user.preference: loves coffee) - Facts are embedded — stored as vectors for semantic search (locally, using fastembed)
- You ask questions —
"what does alice like?"→ returns relevant facts ranked by similarity + recency
Two ways to use it
Personal (MCP)
Plug it into Claude Code, Cursor, or any MCP-compatible client. Your AI assistant remembers things across sessions — project context, your preferences, past decisions.
Multi-tenant (HTTP API)
Build AI products where every end-user gets their own persistent memory. You get an API key per app, your users are identified by any string (user_id), and memories are isolated between users.
Why self-host?
mem0 and Supermemory are cloud services. They work great, but your users' memories leave your infra. For healthcare, legal, and fintech products — or any customer with strict data requirements — that's a dealbreaker.
superdupermemory runs on your AWS, GCP, or bare metal. We can help you set it up.