#135135th of 193 in Skills, plugins and prompts

memsearch

Persistent semantic memory for Claude Code, Codex, OpenClaw and OpenCode agents

Stars
2.8k
License
MIT
Last commit
Oct 2026
Last release
Sep 2026
Language
Python
Stars, last 29 weeks+1,744
31 Mar 2026: 1,023 stars7 Apr 2026: 1,114 stars14 Apr 2026: 1,183 stars21 Apr 2026: 1,255 stars28 Apr 2026: 1,478 stars5 May 2026: 1,573 stars12 May 2026: 1,641 stars19 May 2026: 1,752 stars26 May 2026: 1,815 stars2 Jun 2026: 1,863 stars9 Jun 2026: 1,927 stars16 Jun 2026: 2,027 stars23 Jun 2026: 2,081 stars30 Jun 2026: 2,143 stars7 Jul 2026: 2,188 stars14 Jul 2026: 2,218 stars21 Jul 2026: 2,298 stars28 Jul 2026: 2,379 stars4 Aug 2026: 2,402 stars11 Aug 2026: 2,440 stars18 Aug 2026: 2,475 stars25 Aug 2026: 2,504 stars1 Sept 2026: 2,541 stars8 Sept 2026: 2,571 stars15 Sept 2026: 2,601 stars22 Sept 2026: 2,632 stars29 Sept 2026: 2,676 stars6 Oct 2026: 2,715 stars13 Oct 2026: 2,741 stars20 Oct 2026: 2,767 stars

1,023 on 31 Mar 20262,767 on 20 Oct 2026

From GitHub's daily star counts.

Checked on GitHub.

Zilliz is hiring: 11 open roles

Overview

memsearch captures each agent conversation turn as an LLM-written summary in daily Markdown files, then indexes them in Milvus with hybrid search (dense vector, BM25, RRF). Plugins exist for Claude Code, Codex, DeepSeek Harness, OpenClaw and OpenCode, and a CLI and Python API allow use in custom agents. Recall runs in three layers: search, expand, then raw transcript.

Who it is for: Users of coding agents who want searchable memory across sessions and tools

Strengths

  • Markdown files are the source of truth; the Milvus index is rebuildable
  • One memory shared across five agent platforms
  • Default local ONNX bge-m3 embeddings need no API key; Milvus Lite needs no setup
  • Also exposes a CLI and Python API for custom agents

Weaknesses

  • First run downloads a ~558 MB embedding model from HuggingFace
  • Capture summaries call an LLM on every turn, adding model usage
  • Codex install needs `codex --yolo` for ONNX network access
  • Team use needs a separate Milvus server or Zilliz Cloud

What it needs

  • no GPU
  • Needs Milvus (Lite by default), HuggingFace Hub (model download), Python >=3.10
  • Models: ONNX bge-m3, OpenAI embeddings, Ollama

Also in Skills, plugins and prompts

See all 193
Also in Skills, plugins and prompts
RankProjectScore
1cavemanCuts token use in coding agents with terse replies and compressed inputs Live demo ↗ (opens in a new tab)114k starsApache-2.095 out of 100
2impeccableDesign skill with 24 commands and a detector for AI-generated frontends Live demo ↗ (opens in a new tab)84.1k starsApache-2.095 out of 100
3skillsAgent skills for planning, TDD, debugging and code review299.1k starsMIT94 out of 100
#4eccAgents, skills, hooks and rules bundle for Claude Code, Codex and other harnesses285.1k starsMIT94 out of 100
#5oh-my-claudecodeMulti-agent orchestration plugin and CLI for Claude Code40.1k starsMIT93 out of 100