Supermemory
Supermemory is context infrastructure for AI agents — a five-layer memory stack covering user profiles, a custom vector-graph engine, hybrid retrieval (vector + keyword, sub-300ms), document extractor
About
Supermemory is context infrastructure for AI agents — a five-layer memory stack covering user profiles, a custom vector-graph engine, hybrid retrieval (vector + keyword, sub-300ms), document extractors for PDFs/images/audio, and connectors to Notion/Slack/Google Drive/Gmail. It benchmarks #1 on LongMemEval, LoCoMo, and ConvoMem, and processes 100B+ tokens monthly. There's also a personal consumer app giving users a single memory layer shared across Claude Code, Cursor, OpenClaw, and other tools.
Development teams building production AI agents that need rich, evolving user context — particularly multi-source retrieval, user profile modeling, and long-term memory across conversations — where basic vector stores aren't sufficient.
Pros & Cons
Pros
- check Five-layer context stack in one API — avoids stitching together separate RAG, memory, and extraction services
- check Sub-300ms retrieval at scale, with benchmark-leading results on major memory eval frameworks
- check Memory graph with ontology-aware edges — knowledge updates and infers rather than just appending
- check Connectors to major sources (Notion, Slack, Drive, Gmail) with auto-sync, no manual imports
- check Self-hostable with SOC 2 compliance and TypeScript/Python SDKs
Cons
- close More infrastructure than most small agent projects need — a vector database may be sufficient for simple use cases
- close Pricing for the full API stack isn't immediately transparent on the homepage
- close The consumer app and developer API are distinct products that share a brand, which can be confusing when evaluating
- close Some benchmark claims (especially on their own MemoryBench platform) have self-referential credibility concerns
More Memory
Other tools in the same category.
