A memory/context layer for AI that automatically extracts facts from conversations and builds user profiles, making up for AI's inability to remember information across conversations
Handles updating knowledge, resolving contradictions, and even deleting expired information (automatic forgetting)
Understands that "just moved to SF" replaces "lives in NYC", and expires temporary facts such as "have an exam tomorrow" once the date has passed
Provides Hybrid Search that combines Memory + RAG in a single query, returning knowledge base documents together with personalized context
Automatically maintains User Profiles — provides stable facts (static) + recent activity (dynamic) in a single call (about 50ms)
Connectors automatically sync Google Drive, Gmail, Notion, OneDrive, and GitHub via real-time webhooks
Built-in Multi-modal Extractors — PDFs, images (OCR), video (transcription), and code (AST-aware chunking) are processed just by uploading
Developers can add memory/RAG/profiles/connectors with a single API, with no need for vector DB setup or embedding pipelines/chunking strategies
Provides an MCP server and plugins — supports Claude Code, Cursor, VS Code, OpenCode, OpenClaw, Hermes, and more, with memory/recall/context tools
Ships as a single binary that runs right away on localhost:6767 with no setup, and works fully offline when integrated with Ollama
Provides drop-in wrappers for Vercel AI SDK, LangChain, LangGraph, OpenAI Agents SDK, Mastra, Agno, n8n, and more
Ranked first on three major AI memory benchmarks: LongMemEval (81.6%), LoCoMo, and ConvoMem
Also released its own open-source benchmark framework, MemoryBench