{
  "markdown": "# ZAM (Zusammen) 🤝\n\n**Do real work with your AI — and keep the knowledge instead of losing it.**\n\n> *ZAM is Bavarian for \"together\".*\n\nZAM turns everyday work with your AI agent into active-recall practice, so growing\nautomation doesn't mean growing dependence. You get the task done **and** you get\nsharper — the two stop being a trade-off.\n\nDon't just automate. **Elevate.**\n\n🌐 **[zam-os.org](https://zam-os.org)** — the project website, in 7 languages.\n\n---\n\n## Who ZAM is for\n\nAnyone who works with an AI agent and doesn't want to get rusty as it takes on more —\nwhether you're learning the field (say, a *Fachinformatiker* apprentice building durable\ncompetence), sharpening your craft on the job, or simply keen to keep growing while you\nautomate. If you're pairing with Claude, Codex, Copilot & co. anyway, ZAM makes that time\ncompound into lasting skill.\n\n---\n\n## What ZAM does today\n\n- **Rides along in your AI agent.** As you work a real task, ZAM breaks it into small\n  knowledge concepts, notices which ones you're due to revisit, and weaves them into\n  the session.\n- **Watches you work.** Do a step well on your own and ZAM quietly marks it learned — no\n  interruption. When no real task can show what you know, it asks a focused recall\n  question. Both are active recall.\n- **Remembers what you're forgetting.** Every concept is scheduled with FSRS-6 spaced\n  repetition, including short learning and relearning steps, over a prerequisite graph.\n- **Stays on your machine.** One local SQLite database (`~/.zam/zam.db`), shared by the\n  agent and the Desktop Studio. Review works offline; local LLMs (Ollama, FastFlowLM)\n  are supported.\n\n---\n\n## Two places to use ZAM\n\nYour **agent app** is the main workbench. **ZAM Desktop Studio** is for setup, content,\nand focused review. They share the same local database, so progress in one shows up in\nthe other.\n\n### 1. In your AI agent — *the workbench*\n\nThis is where the real learning happens: turning actual tasks into practice, observing\nyour work, and guiding you step by step. ZAM connects to the agent apps you already use:\n\n| Agent | Connect with |\n|---|---|\n| **Claude** (Code / desktop app) | `zam agent connect claude-code` |\n| **Codex** | `zam agent connect codex` |\n| **Antigravity** | `zam agent connect antigravity` |\n| **OpenCode** | `zam agent connect opencode` |\n| **GitHub Copilot** (CLI / app) | `zam agent connect copilot` |\n| **Goose** | `zam agent connect goose` |\n| **Hermes** | `zam agent connect hermes` |\n\nOne command writes the MCP config (your agent may ask you to approve the server). For\nGitHub Copilot, it also installs user-scoped Studio, Recall, Graph, and Settings canvases;\nrestart Copilot or start a new session after connecting. Then just type **`/zam`** — or\nsay \"let's do this together with ZAM\" — and work normally.\n\n#### Portable Agent Plugin\n\nThe repository and published npm package also follow the vendor-neutral\n[Agent Plugins v1.0.0](https://agent-plugins.org/) format. Compatible clients can load\nthe package root to discover the ZAM skill and stdio MCP server together. A source\ncheckout needs `npm ci && npm run build` first; published packages already contain the\nruntime. See [ZAM Agent Plugin](docs/AGENT_PLUGIN.md) for the layout and validation.\n\n### 2. ZAM Desktop Studio — *setup, content & graph*\n\nA native app (`zam ui`) for the things a chat window isn't good at:\n\n- **Guided setup** — the first start connects your AI model, agent, and workspace one\n  page at a time; everything stays editable in Settings later.\n- **Easier configuration** — pick your language and local AI model in a settings panel,\n  not a config file.\n- **Import your own material** — paste notes, point ZAM at a source, or walk a structured\n  curriculum; a guided wizard turns any of them into review cards.\n- **Edit your content** — a real editor for concepts, questions, and prerequisites.\n- **See your knowledge graph** — your concepts as a living map of what builds on what.\n- **Review** — run focused active-recall rounds right in the app.\n\n```bash\nzam ui            # launch the Studio\nzam ui --build    # one-time: build a native installer (needs Rust)\n```\n\n> **Review works in both places.** Observation and guided task-work happen inside your agent.\n\n---\n\n## Quickstart\n\n**1. Get ZAM.** One line installs the desktop app and the `zam` CLI:\n\n```bash\n# macOS · Linux\ncurl -fsSL https://zam-os.org/install.sh | sh\n```\n\n```powershell\n# Windows · PowerShell\nirm https://zam-os.org/install.ps1 | iex\n```\n\nOr grab an installer from [Releases](https://github.com/zam-os/zam/releases), or build\nfrom source (`git clone` → `npm install && npm run build`).\n\n**2. Open ZAM.** The first start walks you through setup, one page at a time: who you're\nlearning as, your AI model (cloud or local), your agent, your workspace, and your first\nlearning content — from a curriculum, your own sources, a project, or a goal you name.\nEvery page can be skipped and finished later; the dashboard keeps the remaining steps as\na checklist. No terminal required.\n\n**3. Learn while you work.** Open your agent, start a real task, and type **`/zam`**. It\nchecks what's due, plans the concepts behind the task, hands you the work, watches how it\ngoes, and updates your schedule. Prefer a gentler start? Import material and do a review\nround right in the Studio.\n\nMulti-device (a server database and mobile pairing) is an optional later upgrade in\nSettings — the first run stays fully local.\n\n### Prefer the terminal?\n\nThe same setup runs as commands — `zam init` is the guided wizard, CLI-style:\n\n```bash\nzam init                        # workspace · AI model · database · /zam skill\nzam agent connect claude-code   # or codex · antigravity · opencode · copilot · goose · hermes\n```\n\n---\n\n## How it works\n\n- **Token** — one atomic concept worth remembering, tagged with a Bloom level (1 remember → 5 create).\n- **Card** — your personal spaced-repetition state for a token (FSRS-6).\n- **Prerequisites** — a graph of what must be understood first; ZAM won't quiz a concept\n  whose foundations you've just forgotten.\n- **Sessions** — every work/learning episode is logged, so ratings come from real evidence.\n\nThe learning engine is an **AI-agnostic kernel** with zero LLM dependencies; the agent\nlayer just drives it. See [Architecture](docs/ARCHITECTURE.md).\n\n---\n\n## Documentation\n\n- 🌐 [zam-os.org](https://zam-os.org) — project website, in 7 languages\n- [Usage & maintenance](docs/USAGE.md) · [Contributing](CONTRIBUTING.md) · [Architecture](docs/ARCHITECTURE.md)\n\n## License\n\nApache 2.0 — see [LICENSE](LICENSE).\n",
  "bytes": 6586,
  "sha": "9a31d674ae38e2ae562617245e39dde9c993a756c1a5a650af4dfd7e260efa3f",
  "repo_slug": "zam-os/zam",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/okf_zam_os_zam_docs_okf_index_md_319eeb61/readme"
}