memxt
Local memory for AI coding agents
memxt gives AI coding agents a persistent, on-device memory they can search and write to — so your agent remembers your codebase and decisions across sessions. It installs into Claude Code as a plugin (MCP tools + session hooks) and works with any MCP client like Cursor, Zed, or Windsurf. One static binary: no Python, no cloud, nothing leaves your machine. Written in Zig.
Quickstart
curl -fsSL https://raw.githubusercontent.com/Yupcha/memxt/main/install.sh | bash/plugin marketplace add Yupcha/memxt
/plugin install memxt
Why it matters
Most AI coding sessions start amnesiac — you re-explain the architecture and the agent contradicts last week's decision. And most memory layers ship your code to a cloud API. memxt keeps it local.
Remembers across sessions
Semantic recall over everything you've mined or saved.
100% local & private
Embeddings, storage, and search run on-device — no API keys, no network at query time.
Claude Code native
Install as a plugin: MCP memory tools, session hooks, and /remember, /recall. Or point any MCP client at it.
One static binary
~6 MB plus a 45 MB embedding model — no Python, no Docker, no vector database to run.
What's inside
Claude Code plugin
MCP memory tools, SessionStart/PreCompact hooks, and /remember, /recall slash commands.
Any MCP client
Point Cursor, Zed, or Windsurf at memxt mcp.
Local embeddings
llama.cpp + sqlite-vec — sub-millisecond vector search once resident.
Open & MIT
Fork it, embed it, ship it.