Set up without an agent

Some teams may not pass their code through a coding agent: a bank's, often. You can still put Rendel in your app and go live, by hand. Nothing on this page needs a setup file, a setup token or an agent, and the result is the same app the agent would have made. Your developers install the SDK themselves, and the rest is filled in through the console's own forms.

It asks more of a person than Set up with AI, because someone on your team reads the app instead of an agent.

1. Create the app

In the console, Create app, give it your app's name and, if you like, its store listing or website; Rendel reads the name and icon from the link. You land on its Overview. Leave the AI setup card alone: you will not open a setup session.

2. Install the SDK

Copy the test key from Settings → Keys & environments → Copy. A publishable key is public by design: it names your app and nothing more.

dependencies:
  rendel:
    git:
      url: https://github.com/neon-apps/rendel-flutter.git
      ref: v0.48.0

await Rendel.init(
  appKey: 'rd_pk_test_…',
  config: RendelConfig(languages: ['tr']), // or ['tr', 'en'], ['en']
);

<script src="https://rendel.ai/sdk/web/0.25.0/rendel.js" data-app-key="rd_pk_test_…" data-languages="tr"></script>

Run the app once. That is the first two Go live checks: the SDK has called in, and it sent its manifest.

The reference pages say the rest, in the order you will want it: who is signed in (Flutter, Web), actions your app runs, and screens Rendel may open.

3. Fill in the console

Every page an agent would fill in has a form of its own. Each saves a draft your test key uses at once.

• Knowledge: your website, store listing, FAQ or files. One source marked Ready is the third Go live check. See Knowledge grounding.
• Actions: under On server, Connect an API reads your API's address or documentation and lists its endpoints to tick, each with its risk level. Actions your app's code registers appear under In app code after a build. See Connect an endpoint.
• Screens: listed from your build once it registers them; a switch stops Rendel offering one.
• Experience: the assistant's name, tone, rules, languages and welcome screen. See Write the persona.
• Appearance: colours for light and dark, corners, type and logo. See Change the look.

4. Try it

Ask it what your users would on the Test page. Talk as picks a signed-in user, so actions that run as them run for real. Then press Run smoke test on the Overview: when 80% of its answers come from your knowledge or your actions, the fourth check passes.

5. Go live

With the four checks green, Go live on the Overview publishes your drafts and turns the live key on. The checks are the server's, the same ones a setup agent has to pass. Swap the live key into your release build, list your web domains under Settings → Security → Allowed domains, and ship.

From then on, change the console by hand as before, or sync with code if your rules later allow an agent.

What this path does not do

• It is offered, not enforced. Today any member with the right role can still open an AI setup session for the app. A switch that refuses setup tokens and agent connections for a whole organization is planned.
• There is no setup file to write by hand yet. A public schema for one, so a person can write it and import it, is on its way; until then the console's forms are the way.