Oracle agent memory cuts input tokens ~10x at turn 80 vs full chat replay
bibryam · x · 2026-11-18
bibryam details an agent memory architecture built with Oracle: at turn 80, it used 1.3K estimated input tokens versus 13.9K for replaying the full scripted chat — a 10x reduction.
- The memory core extracts durable facts, maintains thread summaries and context cards, and orchestrates retrieval
- The memory store persists raw messages and extracted memories, supporting vector search and retrieval scoped by user, agent, and thread
- The split lets memory policy and storage evolve separately
More from coding & agent
- Telegram preps built-in AI agent "Dictator" on Nous Research's Hermes for 1B+ users — testingcatalog · 2026-11-18
- Google rolls out Antigravity Orchestrator: dynamic in-task model switching, 24% less quota — rseroter · 2026-11-18
- Using Google Antigravity agent swarms to analyze the November 2026 Core Update — gaganghotra_ · 2026-11-18
- Dev builds MCP skills tool so Claude, Cursor and Copilot share one set of instructions — EntrepreneurOk5764 · 2026-11-18
- 12 indie AI mini-game devs share real revenue: top Steam title made $42K RMB in 49 days — sujingshen · 2026-11-18
- Huawei researcher's Cambridge talk: agent-era code bottlenecks shift to specs and verification — satnam6502 · 2026-11-18