Skip to main content

Memory Tree

The Memory Tree is a local, semantic knowledge base that lives on your machine. It stores everything the agent learns about you, your projects, and the world - in a form that can be retrieved efficiently during conversation.

Structure​

Root
 └── Persona
      └── Identity (name, preferences, patterns)
 └── Projects
      └── project_a/
      └── project_b/
 └── World
      └── topics/
      └── contacts/

Storage​

  • SQLite - <workspace>/memory_tree/chunks.db
  • Markdown - <workspace>/wiki/ (bidirectional sync with Obsidian)
  • Attachments - inlined or file-referenced alongside chunks

How memories enter the tree​

  1. Auto-fetch - continuous background sync from integrations
  2. Learning layer - the agent explicitly stores what it learns during conversation
  3. Triggers - incoming events are acknowledged and written as memory notes

How the agent retrieves them​

The recall tool queries by topic, source, or natural language. The retrieval layer returns ranked chunks with provenance.

See also​