Memoryfields: agent memory as portable Markdown files plus an optional SQLite index
rseroter · x · 2026-09-03
Cal Paterson proposes Memoryfields, a radically simple take on agent memory: portable Markdown files plus an optional SQLite vector index. He argues real agents should never start from a blank context window, and critiques three dominant memory designs:
- Vendor-locked systems that mine your conversation history — mostly storing facts about you, when world knowledge is more useful; a play by labs to move from API to platform business.
- Overcomplicated stacks — one prominent system needs pgvector, a Neo4j graph DB, and its own LLM just to decide what's worth remembering; hard to run, confusing to models, and won't scale with the model frontier.
- 'High Modernist' designs with graphs and logical propositions that strip context and leave information isolated and meaningless.
rseroter amplifies it with the question: are we over-processing agent memory?
More from coding & agent
- NVIDIA's NOOA represents AI agents as plain Python objects, ditching prompt/config sprawl — burkov · 2026-09-03
- NVIDIA's NOOA turns agent scaffolding into ordinary Python objects — SimplyAnnisa · 2026-09-03
- Codex computer-use pro-tip: a shared task-title convention to track agent browser tabs — danshipper · 2026-09-03
- Pro tip: use a shared emoji task-title convention to track Codex computer-use sessions across tabs — danshipper · 2026-09-03
- Google ships a Genkit Go skill for agents building AI apps in Go — adonis_singh · 2026-09-03
- Codex Deleted the 6 Test Cases That Uncovered a Bug So Its Tests Would Pass — PuzzleheadedAnt9503 · 2026-09-03