HumemAI is an open source organization building memory systems for agentic AI. We publish research, code and benchmarks for a memory layer that stays persistent across sessions, works across conversations, documents, tables and graphs, and stays inspectable instead of turning into a black box.
- arcadedb-embedded-python: ArcadeDB, a multi-model database engine, embedded in Python. Docs · PyPI
- humemdb: an embedded multi-model runtime with SQL, Cypher, and exact and approximate vector search. Docs
- cypherglot: a compiler that lowers Neo4j-style Cypher into SQLGlot. Docs
- audit-ready-memory: local-first memory for AI agents, with transparent logs, deterministic replay and explicit deletion. Docs
The code behind each paper:
- roomkg-meta-policies: neuro-symbolic meta-policies for temporal knowledge-graph memory under partial observability (ISWC 2026)
- kg-memory-transfer: short-term to long-term memory transfer for knowledge graphs under partial observability (RLC 2026)
- co-learning: knowledge graph-based episodic memory for human-robot teamwork in urban search and rescue (RO-MAN 2026)
- roomkg-baselines: the Room environment v3 and its baselines
- agent-room-env-v2: knowledge graph-based human-like memory systems for partially observable Markov decision processes
- explicit-memory: a machine with short-term, episodic and semantic memory systems
- human-like-memory-systems: a machine with human-like memory systems
- room-env: the Room environments