Skip to main content
Five minutes from an app with no agent to one that answers with a live screen. Vendo brings the agent and its chat surface. You wire a provider and an overlay.

1. Install

2. Run init

It asks a few questions. On the model one take Vendo Cloud: a free key lands in .env.local as VENDO_API_KEY and the agent runs on it. Your own ANTHROPIC_API_KEY works too. Init turns it into the models: line that selects the model, which a loose key never does. Then it writes the server route, your API as tools, and your brand as a theme. Every flag and every file it touches: vendo init. Monorepo? Every vendo command operates on the directory it runs from, so install and run inside the app package (npm install @vendoai/vendo -w apps/web, then init, login, doctor, and sync from there) — details in vendo init. Not Next.js? → Edge runtimes.

3. Mount the provider

Init prints this paste at the end of the run. It never writes a client file, so this step is yours.
The overlay resumes the user’s last conversation across page reloads and offers their previous conversations from a header button — no wiring needed. Pages Router, a prefixed baseUrl, other surfaces: Surfaces.

4. Ask it something real

Open your app, click the launcher pill, and ask for something you would otherwise click through to, like “Show my overdue invoices by customer.” You get a screen back.

5. Check the wiring

Doctor reads your composition, makes one live round trip, and probes that credentials forward on a real tool call. Red output names an E-code. Look it up in Troubleshooting. Next: theme it, register your own components, or deploy it.