apsolut-cortex
Starter Persistent memory for Claude Code — corrections, decisions, and patterns across sessions.Source of truth
| Surface | URL |
|---|---|
| GitHub | github.com/apsolut/apsolut-cortex |
| npm | npmjs.com/package/apsolut-cortex |
Current npm line: 0.13.x (check npm view apsolut-cortex version). Node ≥ 20.
This page is the operator cheat sheet. Deep docs live in the cortex repo (docs/OPERATIONS.md, CONFIG.md, …).
Problem it solves
Claude Code (and most coding agents) forget between sessions. Cortex captures observations as you work, compresses them at session end, and re-injects relevant memory next time — so you stop re-explaining decisions and corrections.
- Storage is local (SQLite-compatible under
~/.apsolut-cortex/) - Compression uses Anthropic API or Ollama (fully local option)
- Optional pair with apsolut-seshat for a curated markdown vault
Install
bash
npm i -g apsolut-cortexIn a project:
bash
cd your-project
apsolut-cortex initRestart Claude Code. Hooks wire memory automatically.
Requirements
- Node.js ≥ 20
- A compression provider:
ANTHROPIC_API_KEYor local Ollama - Windows: hooks run via Git Bash — run
apsolut-cortex doctorif something fails
Everyday (inside Claude Code)
text
/apsolut-recall <topic> # search memory
/apsolut-store <content> # save something explicitly
/apsolut-status # memory stats
/apsolut-forget <topic> # delete a wrong memoryCLI (from shell — confirm with --help on your version):
bash
apsolut-cortex --help
apsolut-cortex doctor
apsolut-cortex status # if available on your versionHow it fits the toolkit
text
seshat → project vault (.apsolut/) you curate
cortex → automatic agent memory across sessions
skills → reusable playbooks (apsolut-skills)
helpers → this site: CLI how-tos| Need | Tool |
|---|---|
| Structured notebook / decisions you own | seshat |
| “Don’t forget what we fixed last week” | cortex |
| Installable agent procedures | skills |
| Command cheatsheets | Helpers docs/ |
Config (high level)
Env vars are prefixed APSOLUT_CORTEX_* — see CONFIG.md in the repo.
Never commit API keys. Prefer direnv / secret managers for ANTHROPIC_API_KEY.
Privacy
- Memory DB is on your machine
- Compression may call Anthropic unless you use Ollama
- Don’t store secrets or client credentials as “memories”