Sync with code
When your app gains a screen, an API call or a new feature, the same agent brings the console up to date — and on a live app, a person reviews every change first.
Set up with AI happens once. Your code keeps moving: a new screen, a new endpoint, a changed theme, a feature the copilot should be able to explain. Sync with code runs your coding agent over the differences, with a second file, rendel-sync.md.
Getting the sync file
Press Sync with code in the console. You choose your agent and get the same two ways to fetch the file as at setup: a one-time curl command for the project root, or a download.
curl -fsSL "https://rendel.ai/api/setup/file?code=<code>" -o rendel-sync.mdThen start it the way you started setup, with rendel-sync.md in place of rendel-setup.md — for Claude Code, claude --settings '…' "Set up Rendel in this project: read rendel-sync.md and follow it step by step, …"; the console shows the whole line. The agent table has the rest.
The sync file carries the same kind of token as the setup file, with the same 24 hours and the same limits (see Token and security), and one difference: it may still talk to an app that has gone live. A setup file cannot.
What the agent does
It reads what Rendel already has — every knowledge source, action and screen with the name it was written under — compares that with the code, and sends only the differences:
| In your code | What the agent does |
|---|---|
| A new screen | Registers it with registerRoute and declares it |
| A new API call | Adds an HTTP action, or registerAction where the call cannot be made from a server |
| Something that is gone | Proposes archiving it. Nothing is ever deleted |
| A changed theme | Sends the new theme |
| A new feature | Updates the knowledge text that describes it, under the same name |
Steps with nothing to change are skipped. Code changes go on a new branch, rendel/sync, committed and never pushed, for you to review like any other branch.
Before go-live: applied, as at setup
On an app that has not gone live, a sync's changes land the way setup's do: Appearance and Experience into their drafts, knowledge and actions directly. Nobody is using the app yet, so there is nothing to protect.
After go-live: proposed, then reviewed
On a live app, the console is somebody's production. Nothing a sync sends is applied. Each change becomes a proposal, and the agent is told how many it made (the API answers 202 with proposed).
The proposals wait on Overview, in a Sync review card, grouped by area: appearance, experience, knowledge, actions and screens. Each shows what it was and what it would become, with the agent's reason. Apply them all, or apply and reject them one at a time. Applying an appearance or experience change still lands in the draft, which you publish as usual.
The code half of a sync — new registerRoute and registerAction calls — is already on the rendel/sync branch, and goes through your own review and release.
Knowing when to sync
You do not have to remember. The console notices drift on its own and marks Sync with code with it:
- A build reports something new. Each time your app starts, the SDK sends its manifest. When a build registers screens or actions the console has not seen, it says so.
- Your API documentation changed. Actions imported from an API reference are watched, and a change to the document is flagged.
Neither needs a file or an agent to notice; running a sync is how you act on them.
Changing the console directly
A sync is a convenience, never the only door. Every console change it proposes can be made on the Knowledge, Actions, Screens, Appearance and Experience pages directly.
Set up with AI, in detail
Which agents run the setup file, what the agent changes in your code and in the console, the token it carries, and what to do when it gets stuck.
Write the persona
The structured persona editor, the three strings the model reads, and what belongs in each: how the copilot speaks, what it refuses and when it points the user to a person.