Quickstart checklist

Work through this checklist while you install CKEditor AI On-Premises. It lists what you prepare, in the order you need it, and it ends with one request that proves the deployment works.

The checklist does not replace the install instructions. Generate the docker run command with the Setup wizard, or follow the Deployment guide step by step, and tick the boxes here as you go. Your browser stores the tick marks, so they survive a reload and nobody else sees them. Select Reset to clear them.

Each box above checks one part. This last step checks that the parts work together. One request uses your token endpoint, the access key, the database, and your LLM provider.

Save the following HTML as document.html.

<h2 data-id="e1">Quarterly product update</h2>
<p data-id="e2">we shipped the new dashbord to all customers this quater. it should of been out sooner but we hit some bugs in the rollout.</p>
<table data-id="e3" class="metrics-table">
	<thead>
		<tr><th>Metric</th><th>Q2</th><th>Q3</th></tr>
	</thead>
	<tbody>
		<tr><td>Active teams</td><td>1,204</td><td>1,688</td></tr>
		<tr><td>Uptime</td><td>99.95%</td><td>99.97%</td></tr>
	</tbody>
</table>
<ul data-id="e4">
	<li>Faster exports
		<ul><li>PDF and Word</li></ul>
	</li>
	<li class="highlight">New review workflow</li>
</ul>
<p data-id="e5"><em>Numbers above are unaudited.</em></p>

The command needs curl and jq. Replace the two placeholders before you run it:

• <your-token-endpoint-url> – the token endpoint you created under Installation above. The token it returns must carry the ai:documents:process permission.
• <your-base-url> – the address CKEditor AI On-Premises listens on, for example https://ai.example.com.

# 1. Get a token from your own token endpoint.
TOKEN=$(curl -s "<your-token-endpoint-url>")

# 2. Build the request from the file and send it to your deployment.
jq -n --rawfile doc document.html '{
	content: [ { type: "document", content: $doc } ],
	prompt: "Fix the spelling and grammar in the introduction paragraph. Leave everything else exactly as it is.",
	model: "agent-1"
}' | curl -s "<your-base-url>/v1/documents/process" \
	-H "Authorization: Bearer $TOKEN" \
	-H "Content-Type: application/json" \
	-w "\n(completed in %{time_total}s)\n" \
	-d @-

A successful call returns the edited document and a summary of the changes:

{
	"documents": [
		{
			"document": "[the edited HTML]",
			"summary": "[what the model changed]"
		}
	]
}

Compare the returned document with the one you sent. The request passed if all of the following are true:

• The introduction paragraph is corrected.
• The table is unchanged.
• The nested list is unchanged.
• The class attributes are unchanged.
• The data-id attributes are unchanged.

If anything else changed, the model edited more than the prompt asked for. Check the prompt field in the request, then the model you configured.