An agentic memory library that structures memories dynamically using Zettelkasten principles: each new memory becomes a note with generated context, keywords, and tags, then is linked to related notes — and adding memories can trigger evolution of existing ones. Published at NeurIPS 2025.
- Storage
- Notes (with structured attributes) indexed in a vector store (ChromaDB), forming an evolving interconnected memory network.
- Retrieval
- Semantic search over note embeddings, following LLM-generated links between related notes for context-aware recall.
- Self-host
- Self-host: moderate
- License
- MIT
- Pricing
- Open source (MIT), free to self-host · Free / OSS
- GitHub stars
- 1,067
- Last release
- —
- Last commit
- 2025-12-12
- First catalogued
- 2026-06-28
Strengths
- Peer-reviewed (NeurIPS 2025)
- Dynamic, self-organizing memory via Zettelkasten-style note linking
- Memory evolution: new entries can update existing notes
Watch out
- Multiple repos exist (agiresearch/A-mem, WujiangXu/A-mem, WujiangXu/A-mem-sys) — confirm the canonical one before promoting
- No tagged releases; research-grade maturity
Best for
- Agents needing adaptive, self-linking long-term memory without fixed memory ops
Benchmark results
| Benchmark | Value | Backbone | Trust | Source |
|---|---|---|---|---|
| locomo | 48.38 accuracy | gpt-4o-mini | Independent | MIRIX (Wang & Chen) — competitor re-run ↗ |
| longmemeval | 55 accuracy | gpt-4o-mini | Independent | LiCoMemory (Huang et al., HKUST et al.) — competitor re-run ↗ |
| locomo | 48.59 accuracy | gpt-4o-mini | Independent | LiCoMemory (Huang et al., HKUST et al.) — competitor re-run ↗ |
Sources
- https://github.com/agiresearch/A-mem (vendor)
- A-mem-sys (production-ready implementation) (vendor)
- A-Mem: Agentic Memory for LLM Agents (paper)
Last verified 2026-06-28 · updated by manual-refresh