jevbooks

← All projects

memsearch

zilliztech/memsearch · Homepage

Persistent Markdown memory for AI coding agents, with optional Jev reranking.

Scoring & Ranking100%Runner-up: Classification & Routing

What it is

MemSearch gives AI coding agents (Claude Code, Codex, DSH, OpenClaw, OpenCode) a shared, persistent memory layer. Memories are Markdown files; Milvus is a rebuildable shadow index. It is for agent users wanting zero-effort recall and developers building memory into their own agents.

How it uses Jev

An optional Jev reranker reorders retrieved Markdown memory chunks. It sends Noul questions about each chunk and sorts them by relevance to the query, using the TypeSafe API with no local model download. Bilingual (Chinese/English) evaluation results are provided.

Primitives:noul

Technique worth stealing

Rerank retrieved Markdown chunks by asking a Jev Noul question per chunk and sorting by relevance.

Try it

memsearch config set embedding.provider onnx; see docs/home/configuration.md#optional-remote-reranking

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricno0.04
  • Shows a System One patternno0.39
  • Handles uncertaintyno0.06
  • Measuredno0.12
  • Runnableyes0.89
  • Worth recommendingno0.31
  • Model replicano0.04
  • Problem scopescore on a 0–2 scale1.52
  • About Jevyes0.86

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 22 Sept 2026.