Don't just automate — elevate

Do real work with your AI — and keep the knowledge instead of losing it.

ZAM turns everyday work with your agent into active-recall practice. The task gets done and you get sharper — automation without the dependence.

What it does today

Learning that happens inside your real work.

Rides along in your agent

As you work a real task, ZAM breaks it into small concepts, sees which are due to revisit, and weaves them into the session.

Watches you work

Do a step well on your own and it's quietly marked learned — no interruption. When no task can show what you know, ZAM asks a focused recall question. Both are active recall.

Catches what you forget

Every concept is scheduled with FSRS-5 spaced repetition over a prerequisite graph — resurfaced right before it would slip.

Stays on your machine

One local SQLite database, shared by the agent and the desktop app. Review works offline; local LLMs are supported.

Two places, one brain

Your agent is the workbench. The Studio is for setup, content & the graph.

Both share the same local database, so progress in one shows up in the other.

In your AI agent

The workbench

Where the real learning happens: real tasks become practice, ZAM observes your work, and guides you step by step. One command wires it in over MCP — then just type /zam.

Claude Codex Antigravity OpenCode GitHub Copilot Goose

Observation and guided task-work happen here, inside your agent.

ZAM Desktop Studio

Setup, content & graph

A native app for the things a chat window isn't good at.

  • Easier configuration — language and local AI model in a settings panel, not a config file.
  • Import your material — paste notes, point at a source, or walk a curriculum; a wizard turns it into cards.
  • See your knowledge graph — concepts as a living map of what builds on what.
  • Review — focused active-recall rounds, right in the app.

Review works in both places.

Quickstart

Three commands to your first session.

  1. Set up — one guided wizard picks a workspace, a local AI runner, and wires the skill.
  2. Connect your agent over MCP with zam agent connect.
  3. Work — start a real task and type /zam. It handles the rest.
zsh — ~/dev
$ zam init # guided setup
✓ workspace · local AI · database · /zam skill
 
$ zam agent connect claude-code # or codex · copilot · goose …
✓ MCP server configured
 
# open your agent, start a task, then:
/zam

How it works

A small, honest vocabulary.

token
One atomic concept worth remembering, tagged with a Bloom level (1 remember → 5 create).
card
Your personal spaced-repetition state for a token, scheduled by FSRS-5.
prerequisite
A graph of what must be understood first. ZAM won't quiz a concept whose foundations just slipped.
session
Every work-and-learning episode is logged, so ratings come from real evidence.

The learning engine is an AI-agnostic kernel with zero LLM dependencies — your agent just drives it. Open source, Apache-2.0.