AI Draft
Generate a first-pass pipeline from plain-language instructions without leaving the editor.
AI Draft is useful when you know the outcome you want, but you do not want to manually assemble every pipeline step from scratch. You describe the transformation, review the generated draft, run it against sample input, and only then decide whether to apply the result.
AI Draft is review-first. It generates a draft pipeline and preview output, but it does not change your JSON until you click APPLY.
What AI Draft does
AI Draft turns:
- your plain-language intent
- the JSON in the active editor panel
into:
- a generated pipeline draft
- a short explanation of what that pipeline is doing
- a previewable output you can inspect before applying
Use it for first-pass pipeline creation when you want to move quickly from “I need to clean and reshape this data” to a concrete transform you can review.
Open AI Draft
Open the panel actions for the JSON you want to transform, then choose Generate With AI.
AI Draft works from the active panel, so the sample input in the modal comes from the JSON you are currently editing.
AI Draft requires an OpenAI key connected in Dashboard -> API Keys. If no key is available, generation will not start.
How the workflow works
Describe the result you want
The Prompt panel is where you explain the transformation in plain language.
Examples:
clean the JSON, keep only customer fields, and prepare it for CSV exportflatten the nested address fields and rename the keys to snake_caseremove nulls, keep active users only, and summarize by region
The same panel also shows Sample Input, which is a read-only preview of the current panel JSON that AI Draft will use as context.
Review the generated draft
The Draft panel shows the generated pipeline before anything runs.
It includes:
- an explanation of the proposed transform
- the generated step list
- a pipeline JSON preview
Use this step to sanity-check the plan before running it.
Run the draft and inspect the result
Click RUN to execute the generated draft against the current inputs.
The Output panel shows:
- the transformed result
- execution errors, if the run fails
- a size warning if the result is too large to apply safely
Choose what to do next
After reviewing the output, you can:
- click APPLY to replace the current panel content with the generated result
- use Copy Pipeline JSON to copy the generated pipeline definition
- use Open as New Pipeline to create a separate pipeline for deeper editing
What AI Draft is best for
AI Draft works best when you want:
- a fast first pass for a multi-step transform
- help turning a plain-language request into a concrete pipeline
- a starting point you can refine in the pipeline builder later
Manual pipeline editing is still the better choice when you need:
- careful step-by-step refinement
- reusable production workflows
- direct control over node layout, connections, and configuration details
If the draft gets you close, open it as a new pipeline and continue refining it in the builder. See Building Pipelines.
Constraints to know about
- AI Draft uses the current panel JSON as sample input for planning and preview.
- Request size limits apply, so very large inputs may be rejected.
- Generated drafts are bounded and validated before they can be used.
- Generated output is never auto-applied; you review it first.
- Output size limits still apply when you try to apply the result back into the editor.
Common errors
Missing OpenAI key
If no usable key is connected, AI Draft shows an error telling you to add one in Dashboard -> API Keys.
Invalid request or oversized input
If the prompt is empty or the sample input is too large, AI Draft rejects the request before generation completes. Reduce the input size or tighten the scope of what you are asking for.
Timeout or model failure
Sometimes generation fails temporarily because the model request times out or the upstream request fails. Try again with a shorter prompt or smaller sample input.
Daily limit or temporary capacity
AI Draft can also be temporarily unavailable because of daily usage limits or platform capacity. These errors are temporary and do not mean your document is damaged.