Linggen Memory

linggen-memory

One memory of you, shared by every agent you use — Claude Code, Codex, OpenClaw, and Linggen's own Ling and Yinyue.

Tell Claude Code something on Monday; Codex knows it on Tuesday. One local daemon, one store on your machine. No SaaS, no API key, no signup.

Not just remember and recall — everything a brain does. A good memory holds some things for a day and others for years, brings back what matters, lets go of what does not, collapses duplicates, merges what belongs together, checks its own work, and settles contradictions when it finds them. That is what ling-mem does.

It works on its own. Install it into the agents you already use and stop there if you like. It is also the memory of Linggen — so if you ever meet Yinyue, the companion there, she already knows you on day one.
Status: stable — latest release 1.13.0. Store schema and the CLI/HTTP/MCP contract are frozen. Prebuilt binaries for macOS Apple Silicon and Linux (x86_64, aarch64; glibc 2.39+, Ubuntu 24.04+). The pre-refactor code-indexing tool is preserved at the v0-legacy git tag.

How it works

Both halves of the loop meet at the same store. The day half runs anywhere the plugin is installed. The night half needs Linggen, which ships the mission scheduler — install it alongside Claude Code or Codex to close the loop.

The parts

- ling-mem — the single binary and local daemon that owns the store (LanceDB + Qwen3-Embedding-0.6B, 1024-dim, multilingual). It performs every mechanical operation: write, search, dedup, archive, export. No LLM runs inside it, and every frontend goes through it, so removing a frontend never loses a row.
- The Linggen engine — memory tools for every agent, the always-on identity block at the top of each session, per-turn recall injection, and the capture protocol…