Skip to content

49. Agentic Memory (A-MEM / Zettelkasten-style)

Mini-Project: Zettelkasten-style Agentic Memory

A Zettelkasten-inspired memory system where atomic notes are linked to related notes, supporting both tag-based lookup and graph traversal — enabling an agent to discover non-obvious cross-domain connections when answering questions.

View on GitHub


What It Is

Agentic Memory (A-MEM) applies the Zettelkasten note-taking methodology to agent memory. Each memory is an atomic "note" with rich metadata (tags, links to related notes). Notes are densely interconnected, forming a knowledge graph where ideas link to related ideas. The agent traverses these links to discover non-obvious connections -- creating a "second brain" that grows organically.

What It Stores (Examples)

  • Atomic insights extracted from conversations
  • Linked notes connecting related concepts across domains
  • Source attributions and confidence levels
  • Progressive refinement as notes are updated

How It's Implemented

Uses a graph structure where each node is a note (with embedding for semantic search) and edges represent explicit links.

Mermaid Diagram

flowchart TD
    A[New Info] --> B[Extract Atomic Notes]
    B --> C[Note 1]
    B --> D[Note 2]
    C -->|Related| E[Existing Note X]
    D -->|Supports| F[Existing Note Y]
    E -->|Linked| F
    G[Query] --> H[Vector Search + Graph Traversal]
    H --> C
    H --> E

    style C fill:#2196F3,color:#fff
    style D fill:#2196F3,color:#fff
    style E fill:#9C27B0,color:#fff
    style F fill:#9C27B0,color:#fff