The product

A package is a versioned, named bundle of AI workflow assets, identified as name@version: instructions, skills, agents, workflows, prompts, hooks, MCP servers, provider configs, templates, policies, settings effects and plugin refs. Every asset has one trust class.

`wl push --draft` creates a draft version from selected package assets, a registry admin approves versions, and packages are published on a release channel.

A workspace is the unit that receives packages and gets rendered: a repository, a folder, a monorepo target or your home directory. A home workspace projects AI provider config into your home directory, and `wl install`, `wl upgrade` and `wl uninstall` manage it like a package manager.

Each workspace is described by a workloom.workspace.json manifest with its binding (git, folder, key or home), targets and policies.

The render plan is the deterministic set of files and settings effects for one workspace revision. Sync writes only approved, Workloom-owned projections, after the trust gate and any required local approval pass, then updates the ownership baseline.

Files that existed before are adopted only with --adopt, after a backup. `wl unbind` hands everything back.

The registry API (Hono) holds packages, workspace revisions, approvals, the audit log, webhooks and the public catalog. The dashboard is where you browse the registry and catalog, manage workspaces, read telemetry and administer role mappings. The `wl` CLI does everything on your machine.

The desktop app (Tauri) is, today, a sign-in shell: a device-code login wrapped around a bundled Node sidecar. Workspace previews are not in it yet.