What is a Vibe?
Your AI builds it, we host it.
Last updated: September 2026
A web app built with your coding agent
A vibe is a web app hosted on vibe-coded.ai. You install the vibe-coded.ai CLI skill for the coding agent you already use, describe the app you want, and keep the source in a local project. Your AI writes the code; the CLI runs the worker and its storage locally and sends the source to the platform, which builds it, when you are ready.
The usual workflow is vibe init, vibe dev, vibe preview, then vibe publish. A vibe can be a static site, a worker only (MCP tools and storage, no page), a worker with a page, or a Vue and Vite frontend with a worker. You can also bring an app you already have.
URL structure
After you publish, the production app uses your user slug and vibe slug:
https://{user-slug}--{vibe-slug}.vibe-coded.ai
With the placeholder username you and a vibe named my-vibe, that is:
https://you--my-vibe.vibe-coded.ai
vibe preview deploys the same source to a separate preview URL:
https://p--you--my-vibe.vibe-coded.ai
A vibe with MCP tools exposes them at the app URL plus /mcp.
What the platform provides
- Hosting — Preview and production deployments on a managed runtime.
- Storage —
env.VIBE_STORAGEkey-value storage on every plan. SQL throughenv.VIBE_D1and files throughenv.VIBE_R2are available on Pro and Pro+. See Plans & Billing for current prices and limits. - MCP tools — Only async function declarations marked with
@mcp-exposebecome tools. Other backend functions stay internal. - OAuth 2.1 — AI clients can sign in and receive authorized access to the vibe's MCP endpoint.
- Sharing — Access is assigned per vibe with owner, admin, writer, and viewer roles. Viewers run in read mode: their tool calls can't write to the vibe's storage, though a tool that calls an outside service still runs. Writers, admins, and owners can write. Permissions apply to the vibe as a whole, not individual tools.
What to tell your AI
Describe the data before the layout. Name each record, its fields, allowed values, and relationships. Then say which operations should be available as AI tools.
Build a task tracker with projects and tasks. A project has a
name and color. A task belongs to one project and has a title,
due date, priority, and todo/in-progress/done status.
Expose tools to create and list tasks, move a task between
statuses, search by keyword, and summarize overdue tasks.
If you have a preferred layout or storage type, include that too. Your coding agent can choose the matching scaffold and keep the worker, frontend, and tool definitions consistent.