Agentic Vibe is the open-source toolkit for building apps on top of the ibl.ai platform. You do not build a backend and you do not manage API keys: login, agent chat, profiles, analytics, memory and billing are already wired, and the app is yours from the first commit. The product page is at Agentic Vibe; the developer guide is at Build on ibl.ai.
In this recording, Jaione Amigot builds a customer service app from one prompt. No code is written by hand. She points Claude Code at the iblai/vibe repository, describes the app in a few sentences, and about ten minutes later SupportDesk is running, signed in through ibl.ai and talking to the support agent that already lived in her tenant.
What the demo shows
- 0:00 — One prompt, zero code. The iblai/vibe toolkit, with the backend and login already wired.
- 0:24 — Skills that teach your coding agent to build on ibl.ai. Instruction files installed with one command: agent chat, agent settings, memory, tools, evals, and shipping to the app stores. The full catalog is at Skills.
- 0:43 — Describe the customer service app to Claude Code. SSO login, an agent chat on the home page, a help-desk look, suggested prompts such as "Where's my order?", and conversation history in a sidebar.
- 1:16 — Scaffolding the app and connecting it to your ibl.ai organization. Claude Code reads the README, pulls in the skills, scaffolds the Next.js app and connects it to the tenant.
- 1:51 — Sign in to ibl.ai and pick the tenant. The connect skill opens a browser sign-in that writes the organization key and a minted API token into the app's environment files. No token ever appears in the chat.
- 2:08 — The support agent already in the tenant attaches itself. The agent Jaione had created earlier is recognized and wired to the new app.
- 2:37 — SupportDesk running, one prompt later. Help desk, agent chat, conversation history, and the admin surfaces for users, analytics, billing and memory.
How it works
The skills are instruction files that teach a coding agent how to build against the platform. They cover the pieces every app needs: connecting the project to an organization, SSO, the in-process agent chat, projects, profiles, accounts, analytics, notifications, invitations, workflows, local models and more. You install them once and the agent composes them to order.
What the agent installs is ordinary: Next.js, TypeScript, Tailwind and the ibl.ai SDK. What it does not have to build is the hard part. Authentication, the agent runtime, memory, analytics and billing come from the platform, so the prompt describes the app rather than the infrastructure.
One detail worth pausing on: the app is connected to the organization through a one-click browser flow, and the credentials land in environment files the agent can use but never reads back into the conversation. Register a user on the platform before you start, as the demo notes, and the sign-in step does the rest.
Run it where you want
The demo runs against the hosted ibl.ai backend. The same app runs against ibl.ai on your own servers, where any model answers, your data is reached in place behind your own firewall, the code is yours and every action is logged. Syracuse University runs ibl.ai that way; the university's own account is at ai.syracuse.edu.
Try it
- Clone the toolkit: github.com/iblai/vibe, MIT-licensed.
- Browse the skills: /developer/skills.
- Try it free: ibl.ai/join.
- Book a pilot: Contact us.