easymem

An LLM wiki your coding agent writes — agent memory you can open and read.

English · 简体中文

easymem turns what your coding agent reads into a local wiki of markdown pages, so the next session searches the wiki instead of reading the files again.

Search is BM25 over every page, then one hop along [[links]] — a page that never mentions your words still comes back when a page that matches links to it. Everything lands in plain markdown you can read, edit and commit. There is no model inside easymem and no database: the agent writes, easymem stores, searches and links.

Quick start

With your AI

npx skills add zhangxiangliang/easymem

That installs the skill into Claude Code, Cursor, Codex and other agents. From then on the AI writes down what it reads and searches the wiki first, instead of reading your files again.

No CLI? Hand your AI this line instead, and it does the rest:

Read and follow https://github.com/zhangxiangliang/easymem/blob/main/SKILL.md

On the command line

No install needed. npx gets the package on first run.

npx easymem search "how does checkout work"
npx easymem --help

As an MCP server

The index stays warm between calls. Add this and restart:

{
  "mcpServers": {
    "easymem": { "command": "npx", "args": ["-y", "easymem"] }
  }
}

Pages land in .easymem/wiki/. Commit those; add .easymem/.state/ to .gitignore.

Docs

• Reference — every command, what search returns, what lint checks, what is on disk, and what easymem does not do.

Where this comes from

The idea is Andrej Karpathy's LLM wiki: stop running RAG over raw sources on every question — have the agent compile what it reads into a wiki, then answer from the wiki. I ran that pattern in my own knowledge base for a long time. It works, it just does not work well: with nothing but a prompt, the agent writes each page in whatever shape it feels like that day, and a wiki where every page is shaped differently is a pile.

TencentDB-Agent-Memory is the same idea with real constraints around it, and the results are a lot better. It is also heavy — an LLM client, an HTTP API, a control panel, multi-tenant storage, a SQLite index. easymem lifts out its MemoryKnowledge engine — the mixed Chinese/English tokenizer, the multi-hop graph search, the page format — and puts it behind a shell command. Nothing else.

License

MIT. TencentDB-Agent-Memory is MIT too.