Coding agents forget you between chats. Preferences, decisions, and “don’t do that again” notes vanish unless you paste them back in.
I wanted something simpler than spinning up Docker, Postgres, and an embedding stack just so an agent can remember my name.
That’s amem: local agent memory over MCP. Facts live in ~/.amem on your machine. Free to start. No telemetry.
The problem
Every new agent thread starts cold:
- Who am I building for?
- What did we already decide?
- Which tools are connected?
You either re-explain, or you trust a cloud memory product with your context. Neither felt right for day-to-day Cursor / Claude / MCP workflows.
What amem does
amem is a local memory layer your agent can call:
-
amem_context— pull compact relevant facts before exploring or prompting -
amem_remember— store one durable sentence after a real outcome (preference, decision, recipe)
Memory stays on disk. You can browse it in a local UI.
Quick start
Node 20+:
bash
npx @iamem/amem setup
or globally
npm i -g @iamem/amem && amem setup
then
amem ui or amem app
Top comments (0)