Markdown-file agent memory saves ~155k tokens while injecting only ~12k
gregbarbosa · x · 2026-10-07
Developer gregbarbosa describes a memory system for his work agents built without RAG — just simple markdown files.
- Injects at most 12k tokens into the context window per run
- Yet spends 155k fewer tokens than an agent running without memory enabled
- Precise md memory files replace heavy retrieval, cutting cost while staying focused
Related event: No RAG Needed: Markdown Files Cut Agent Memory Tokens by 155K(3 posts)→
More from coding & agent
- Tip: force GPT-Live to say exact phrases via a developer message in session.start's input — pbbakkum · 2026-10-07
- Dev vibe codes open-source Rust replacements for 7 Adobe apps — alexcovo_eth · 2026-10-07
- Decagon open-sources PACT, a consent & trust protocol for personal AI agents built on A2A and OAuth — kimberlywtan · 2026-10-07
- Every builds a company-wide Slack agent on Claude Managed Agents, then ships it as a product — danshipper · 2026-10-07
- Worse Together: arXiv paper dissects why multi-user multi-agent teams collapse across 77 scenarios — rohanpaul_ai · 2026-10-07
- Stanford paper: one agent per user underperforms a single shared agent on common resources — rohanpaul_ai · 2026-10-07