zilliztech/memsearch
Cross-platform semantic memory for AI coding agents — automatic capture, pre-step memory injection, skill-based recall and background maintenance that flow across Claude Code, Codex, DeepSeek Harness, OpenClaw and OpenCode, backed by Markdown as the source of truth.
MemSearch is cross-platform semantic memory for AI coding agents. Installed as a DSH plugin it brings automatic capture, pre-step memory injection, native skill-based recall, background maintenance and a read-only memory browser to DeepSeek Harness, and the same memories are searchable from Claude Code, Codex, OpenClaw and OpenCode — a conversation in one agent becomes context in the others with no extra setup. It captures and stores memories in Markdown, distinguishes durable PROJECT.md and USER.md notes it keeps current across sessions, and now distills repeated workflows into reusable installable agent skills as a third, procedural memory layer. Agent developers get the full CLI and Python API to build memory and harness engineering into their own agents.
Install
dsh plugin --profile web add @zilliz/memsearch-dshCopied verbatim from the README's DeepSeek Harness platform section. The plugin is one of five first-party platform integrations (Claude Code, Codex, DSH, OpenClaw, OpenCode) that share a single memory store.
Compatibility
Python 3.10 or newer for the library and CLI; cross-platform. Markdown is the source of truth, so memories stay readable on disk. An optional Jev/TypeSafe remote reranker can be enabled with no local model download.
Details
- Repo: zilliztech/memsearch
- Category: Memory
- Stars: 2730
- Version: GitHub repo zilliztech/memsearch; DSH plugin npm package @zilliz/memsearch-dsh at 0.1.6 (registry check 2026-10-09; registry repository field resolves back to this repo). Maintained by Zilliz (the team behind Milvus); the underlying package is published on PyPI as memsearch for Python 3.10+.
Alternatives
clouwer/dsh-memsearch · Amengclass/dsh-memory · Aik358/dsh-auto-memory