// the find
supermemoryai/supermemory
Memory and context engine + app that is extremely fast, scalable, and can be run fully locally. The Memory API for the AI era.
Supermemory is a memory-as-a-service layer for AI agents — it extracts facts from conversations, maintains user profiles, handles contradictions and temporal expiry, and returns relevant context via hybrid RAG+memory search. It targets developers building agents who don't want to roll their own vector DB pipelines, chunking strategies, and memory update logic. Claims #1 on three major AI memory benchmarks (LongMemEval, LoCoMo, ConvoMem).
The static/dynamic user profile split is genuinely useful — static facts ("prefers Vim") and dynamic context ("currently debugging auth") need different retention logic, and they've codified that distinction in the API. The local mode uses the same API surface as the hosted platform with just a baseURL swap, which is rare and makes local dev actually match production. AST-aware code chunking for source files is a real differentiator over generic text chunkers that split on byte boundaries. The temporal forgetting model ("I have an exam tomorrow" expires after the date) is the right approach — most memory systems accumulate noise forever.
The actual memory engine source isn't in this repo — what's here is the app, MCP server, browser extension, and docs. The self-hosted binary installs via `curl | bash` with no auditable source for the core engine, which is a hard no for anything security-conscious. Benchmark claims are self-reported and the MemoryBench framework is their own tool comparing against competitors they chose — treat those #1 rankings as marketing until independently replicated. Container tags as the only multi-tenancy primitive is fine for simple user scoping but falls apart if you need hierarchical permissions or cross-user queries without separate tag namespaces.