Skip to main content
Every Vybe app starts from the same starter project, so all apps share one stack. You don’t need to name technologies in your prompts, because the AI already knows them. Knowing the stack helps when you read the code or ask for something specific. New apps start with the versions below. Apps created earlier may use older versions.

Core

User interface

Data and forms

Collaboration and rich text

Liveblocks is set up for you. There’s no API key to add, and each app’s collaboration rooms are kept separate from other apps. Shared rooms are open to everyone who can use the app, and the AI can make a room private to one person for personal data.

Database

Each app has its own PostgreSQL database. See App data.
  • The AI creates and changes tables by running SQL directly. The app doesn’t use an ORM: prisma/schema.prisma is a locked placeholder.
  • The app’s server code queries the database with SQL through a helper in src/server-lib.
  • Apps can also read and write your organization’s shared databases, and query external PostgreSQL, MySQL and Redshift databases your organization has connected. See Databases.

Background jobs

Apps run scheduled and background work with Inngest. Jobs live in src/inngest/functions. See Background jobs.

Sign-in and users

Vybe handles sign-in, so apps have no login page of their own. Everyone who opens a published app is signed in to Vybe, and the app can read the current user’s name, email and profile image. That’s enough to build per-person features like “show my tasks”. See Access and sharing.

Connections and AI coworkers

From the app’s code, the AI can use:
  • Integrations your organization or you have connected. See Integrations.
  • Custom APIs. Vybe adds the API’s credentials to each request, so they never appear in the app’s code.
  • Data queries.
  • AI coworkers linked to the app, called from server code to get a structured judgment back. See Apps overview.

Built-in integrations

These work in every app without connecting anything:

Testing

The AI writes unit tests with Jest and runs TypeScript type checks after its changes. See Working with code.

Packages

The starter project already includes everything listed on this page. You can’t add arbitrary npm packages: neither you nor the AI can run npm install, and package.json is locked. The AI can install these extra packages when a feature needs them: Ask for the feature and the AI installs what it needs:
Create a workflow builder where people add steps, connect them with arrows and configure each step.
If you need something the stack doesn’t cover, describe what you’re trying to build. The AI can often get there with what’s available, for example by drawing a custom visualization with SVG instead of adding a charting library.

Code conventions

The AI follows the Next.js App Router conventions:
  • Each folder in src/app is a URL path, with page.tsx for pages and route.ts for API routes.
  • Components render on the server unless the file starts with "use client", which interactive components need.
  • Route parameters, like the id in /customers/[id], are read by awaiting params: