Your customer types or says what they want. Rendel finds the right action in your app, asks for anything missing, waits for their tap, then does it through your own API.
Your customer asks
Cancel order 1042, I don't need it anymore.
A customer says what they want. Rendel works out which of your actions fits, asks for anything missing, gets their confirmation when the action changes something, runs it through your existing systems as the signed-in user, and answers with a page built from the result.
Every turn runs the same loop, in the same order: understand, select, ask, confirm, execute, answer with a page. Understand reads the question against the actions you have switched on. Select picks one. Ask fills in whatever the question left out. Confirm is where anything that changes data stops.
Execute calls your API, and answer draws the result as a screen of your app's own components rather than a paragraph about it.
You reuse the logic you have already built and tested. AI setup finds the calls your app already makes and lists them as actions, and you can import API docs or a Postman collection for the rest.
Your team switches on the ones it trusts. An action that is switched off is refused at runtime, without a redeploy.
Anything that only reads runs at once. Anything that changes something runs as the signed-in user, and only after they tap a confirmation card — and that card is built by Rendel's servers from the action's own schema, never from model output.
Every invocation is on the record in the console, per conversation.
The model sees your action names, their descriptions, their parameter types, and the app state you choose to pass.
It does not see your database, your keys or anything you did not hand it.
Every action still passes through the backend APIs, business logic, authentication and permissions you already run.
Rendel does not get a second route into your data; it asks your app the way your app's own screens do.
No. It calls the endpoints your app already uses, as the signed-in customer.
Two things. An action your team has not switched on is refused at runtime, and anything that changes data waits for a confirmation card the customer has to tap.
Yes. Every invocation is recorded in the console against the conversation it belongs to.
Start free on your own, or tell us what your customers do most and we’ll show you that exact workflow, working, in a tailored demo.
Prefer email?
sales@rendel.aiTell us about your app and we’ll show the right workflow in a tailored demo.
