Test as a real user
Create a test user from the Test page and talk as them, or pair a debug build and watch it from the console.
Create a test user
On the console's Test page, open Talk as and choose Create a test user. The console signs a new user up through your app's own sign-up, with a made-up rendel-test+…@test.rendel.ai address, and talks as them at once. Actions that run as the signed-in user, writes included, then run for real against your API.
It is offered when your API document, or what the setup agent reported about your sign-in, has a sign-up and a password sign-in. Otherwise Talk as can add an existing test account, or take a token you paste for the session when sign-in goes through Google, Apple or SSO. The token never reaches the browser, the model or a table.
Pair a debug build
The Test page cannot run handlers that live inside your app. A build of your app can, and pairing lets the Test page watch it.
You want to watch a real build's conversation, with its in-app actions, from the console.
Let me pair a debug build of this project with my Rendel (rendel.ai) console. I am the developer and I am asking for this. In a debug-only developer menu or screen (create a small one if the app has none, never in release builds), add Pair with Rendel: a field for a code like K7Q2-M9XD and a button that calls Rendel.pair(code) and shows the result's message. Rendel.pair never throws and only works with the test key, so nothing changes for release builds. How it works: https://rendel.ai/docs/tasks/test-as-a-real-user Build the app and tell me where to find the menu.
Paste into your coding agent, opened in your project. Works in Claude Code, Cursor, Codex, Windsurf, GitHub Copilot and Gemini CLI.
How to check it worked
On Test, press Pair a test app and type the code it shows into the new menu. The Test page says Paired with your test app and shows what the build is doing: what was asked, what the copilot drew, and each action with its risk and status.
Reference: what it writes
final result = await Rendel.pair('K7Q2-M9XD');const result = await Rendel.pair("K7Q2-M9XD");result.ok is true with endsAt when it worked. Otherwise reason says why (test_only, not_found, rate_limited or failed on the web; RendelPairFailure.testOnly, .notFound, .rateLimited, .failed in Flutter) and message is a sentence you can show as it is.
A code works once, for ten minutes; a pairing lasts two hours or until End pairing. Only a build on the test key can pair. Owners, admins and developers can start one.
The console sees the conversation as it is stored, with personal data already masked, and each action's name, risk and status. Never the user's token, the action's parameters or what your API returned. It watches and sends nothing; confirmations happen on the device.