Lastmemory — AI skill for Claude Code

Persistent, token efficient memory for coding agents.
A network of interconnected markdown files that remembers your project, anchored to the parts of your repo, and forgets what stops mattering.

---

What it is

`lastmemory` is a memory skill for Claude Code, and for any CLI agent that can read markdown. It keeps a small `/memory` folder inside your project where knowledge lives as a network: each file is a neuron (a session, a zone of the repo, a decision, a failure, a reflection), and the links between files are synapses. The agent walks the links instead of loading everything, so it can recall a whole project history while paying for only a few hundred tokens.

It is built on four ideas:

1. Spatial, not chronological. Knowledge is anchored to parts of the repo (auth, payments), not to a flat log.
2. A neural network of files. Every file is a neuron; links are synapses. The agent navigates, it does not dump.
3. Minimum token cost. Layered loading (critical facts, then the index, then summaries, then full content). The whole context is never poured in. This is the headline, and we measure it.