Using Jev, the 'System One' classifier, to speed up self-hosted agent memory
FUNdationOne · reddit · 2026-09-20
The author is building Memry, a free open-source self-hosted agent memory system that keeps your lifetime of memories on your own SSD instead of OpenAI/Anthropic's servers. This week they started integrating Jev — a 'System One' model that classifies and routes instead of generating text, two orders of magnitude cheaper and faster — and report it speeds up entity merging, tagging, decay assertions, and retrieval reranking, with confidence scores enabling clear decision thresholds. Details in their Medium write-up.
More from coding & agent
- Mnemos MCP nears launch after shared memory layer fix — RileyRalmuto · 2026-11-04
- Moving mobile dev to remote machines makes it multiplayer across product, design, engineering — rudrank · 2026-11-04
- Kuaishou's EvoRec: self-evolving agentic recommender lifts ad revenue 4.6% — _reachsumit · 2026-11-04
- TRIO: three specialized agents with aligned objectives refine retrieval-augmented generation — _reachsumit · 2026-11-04
- How Multica uses Multica to build itself: full human-agent dev workflow — jiayuan_jy · 2026-11-04
- Claude Code fixes game copyright refusal rule in 2.1.317: naming a game now works — apples_jimmy · 2026-11-04