Overview
Building a checklist manually is fast, but describing what you need in plain English is faster. The AI template builder converts a natural language description into a valid TDL JSON template — with sections, step types chosen appropriately (boolean for yes/no, number for readings, photo for evidence), required flags, and photo overlay config where needed.
How it works
The AI template builder runs server-side. Your prompt is processed by Rundun's AI layer, which generates a structured template using strict output constraints that guarantee valid TDL JSON. The response is validated against the TDL schema before being returned to you.
Structured output mode guarantees the response is parseable JSON that matches the TDL schema — there's no risk of the model returning prose instead of JSON or omitting required fields.
Using the builder
Via the dashboard
In my.rundun.app, go to Templates → New → AI generation. Type a description of your process and click Generate. Review the result in the visual editor and make any adjustments before saving.
Via the API
POST /v1/templates/generate
Authorization: Bearer rdk_live_...
Content-Type: application/json
{
"prompt": "STR turnover checklist for a 2-bedroom apartment. Include room-by-room checks with photo evidence of each room, linen replacement steps, inventory check, and a final sign-off."
}Response is the complete TDL JSON, ready to save:
{
"name": "STR Turnover — 2 Bedroom",
"sections": [
{
"id": "g-a1b2",
"title": "Bedroom 1",
"steps": [
{
"id": "s-c3d4",
"type": "boolean",
"label": "Bed made with clean linens",
"required": true,
"photo": { "min": 1, "max": 3 }
}
]
}
]
}What the AI gets right
The model is prompted to follow TDL conventions: use boolean for yes/no questions, choice for multiple options, number for readings with units, photo for evidence-only steps. It sets required: true for critical checks and adds photo overlays where visual evidence is genuinely needed.
What to review after generation
Always review AI-generated templates before dispatching to real executors. Check:
- Step labels are specific and unambiguous
- Photo requirements (min) are set correctly — the AI is conservative
- Step order makes physical sense (front to back, room by room)
- Required flags match your actual enforcement needs
Modifying existing templates
The API supports modification prompts on existing templates. The model preserves all existing IDs (critical for any runs already dispatched against this template) and only changes what you ask:
POST /v1/templates/t-550e8400-.../generate
{
"prompt": "Add a bathroom section after the kitchen with 3 steps: check cleanliness, replace towels, confirm mirror is clean with photo."
}Enabling and disabling
AI generation is available on all plans including PAYG. It consumes no run credits — only template creation is involved. You can disable AI generation for your organization from the dashboard settings if you prefer manual-only template creation.