415.tech
AI & tech, from the frontlines of Silicon Valley
Cal Paterson's Memoryfields turns agent memory into a ZIP of markdown pages

Cal Paterson's Memoryfields turns agent memory into a ZIP of markdown pages

Cal Paterson proposes Memoryfields: markdown pages with optional YAML frontmatter and a SQLite vector index, shipped as a ZIP. It drops the pgvector-plus-Neo4j memory stack for semantic search over whole prose pages, cutting retrieval to at most two tool calls. The RFC-style spec, a CLI (memoryfield-tool), and an agent skill are on GitHub, so a developer can swap a vendor-locked memory store for a portable file today.

Source: calpaterson.com ↗

Post on XEmail

At most 2 tool calls are required (#1 to search, #2 to read in parallel). Relevant stuff actually gets found and irrelevant input tokens are minimised.

Cal Paterson

Why this matters

  • → Cuts agent memory retrieval from N+1 tool calls to 2, enabling faster semantic search over prose
  • → Replaces complex pgvector + Neo4j stacks with portable ZIP + SQLite, shipping today
  • → Lets agents invent their own access patterns instead of navigating rigid API mazes
Memory as data, not pipeline