Getting Started
Memory STAK is currently in beta. The CLI and Cloud model are actively evolving.
Introduction
STAK gives you portable Skills, private Memory, and repeatable setup across the AI coding agents you already use. Instead of re-teaching each agent your conventions by hand on every machine, you capture that setup once and bring it with you — to a new machine, a new project, or a different agent entirely.
What is STAK?
STAK is built from a small set of ideas:
- Skills — individual capabilities an agent can use (instructions, context, tools).
- STAKs — named, versioned bundles of Skills you can add to a machine or share publicly.
- Memory — private, Cloud-backed context for an agent that is never included when you publish a STAK.
- Projects — a stable logical identity in STAK Cloud; the local directory on your machine is just that machine's binding to it.
- Machines — the individual computers where you run agents and reconcile Skills.
See Core Concepts for the full detail on each.
Install STAK
Install the CLI globally, then sign in to STAK Cloud:
npm install -g @memstk/cli@beta
stak loginDaily workflow
Once you have Skills and Memory worth keeping, the everyday loop is three commands:
stak status
stak add .
stak pushstak status is read-only and shows what changed. stak add . stages the Skills and Memory you want in main. stak push synchronizes staged main Skills and Memory to STAK Cloud — nothing unstaged is pushed silently.
Move to another machine
On a new machine:
npm install -g @memstk/cli@beta
stak login
stak pull
stak statusstak pull restores main, Skills, and your latest staged private Memory generation. If you use Claude Code, Memory may still need materializing into it:
stak project
stak memory restore --dry-run
stak memory restoreA local-only Project (no Git identity to reconcile automatically) may need an explicit stak project map <project> <directory>. See Memory & Projects for the full explanation, and Public Beta for current limitations.
Your first STAK
The fastest way to try a STAK is in an isolated, temporary runtime that never touches your normal agent setup:
stak try @stephentony100/frontend-test -- claudestak try launches Claude Code with that STAK's Skills projected into a temporary environment. Your real main desired state, your normal agent bindings, and any permanent agent state are untouched — the trial runtime is cleaned up afterward.
If you want to keep using a STAK going forward, add it to main and then launch an agent against it:
stak use @stephentony100/frontend-test
stak run codexThese are two distinct steps with two distinct effects. stak use only resolves and verifies that STAK's Skills and adds them to your main desired state — it does not launch an agent. stak run codex (or claude, or cursor) is the step that projects main into that agent's Skill directory and launches it. See CLI Reference for every command and flag.
Using STAK with Codex, Claude Code, and Cursor
Codex, Claude Code, and Cursor are first-class, equally-supported agents in STAK — none is a primary integration with the others bolted on.
Codex — stak run codex reconciles main into Codex's Skill directory and launches the Codex CLI on your PATH.
Claude Code — stak run claude does the same for Claude Code's CLI.
Cursor — stak run cursor does the same for Cursor's agent CLI.
All three also work with stak try <stak> -- codex|claude|cursor for isolated trials. See Agents for per-agent setup and authentication notes.