# Core > People at work are already building their own software with AI. Core gives them somewhere it's safe: a complete set of work apps (Mail, Calendar, Tasks, Docs, Files, Clients, Projects, Invoices, Data) that is good to work in from day one, and that anyone can make their own by asking, on the company's real data. The apps can't keep or send a copy of that data. One shared source of truth for the whole team (the kernel), the apps on top, and an agent (a chat) that is the main way in: it answers questions about the work, updates records, changes or builds apps for you and brings any part of any app into the conversation. Because apps can't store or send anything, there is exactly one copy of the data, and every change is attributed, logged and undoable. Who it's for: the people who use it get their own tools, on real data, without waiting for IT; the people who sign off (CIO, CTO, COO) can say yes to every tool their team wants to build and keep one source of truth, one permission model and one audit log. Status: a working demo (one workspace, shared demo passcode, no SSO), self-hostable. Key facts: - Three layers, with fixed colours: the kernel (blue), your apps (amber), the agent (violet). - The kernel: types, records, permissions and an audit log in SQLite. 11 field types (text, longtext, number, money, date, datetime, select, bool, ref, account, accounts). Every write is validated against the schema, checked against the writer's role, kept referentially intact, logged with before/after and `via` (which app and version, the agent, or the kernel UI), and can be undone from the log. - Apps: one HTML file each (a JSON manifest, markup, style, script), run in `