Scaffolding with AI

Eddy can generate a draft Map from a text description. Describe the process you want to build, and Eddy produces a complete scaffold — stages, blocks, roles, transitions, and data bindings — that you can review, refine, and open in the builder.

This feature requires an LLM connection configured in your workspace settings.

Prerequisites

A workspace admin must connect an AI provider before scaffolding can generate results. Go to Workspace Settings → AI Connections and add a connection for OpenAI or Google Gemini.

The "Scaffold with AI" tab always appears in the Map creation dialog. If no connection is configured, the tab shows a message and a button that takes you directly to workspace settings to set one up.

Creating a Map with AI

Open the New Map dialog from your workspace. You'll see a tab called "Scaffold with AI".

Writing a prompt

Describe the process you want to build. Be specific about:

• What the process is — "Employee onboarding for new engineering hires"
• The major steps — "intake form, equipment setup, manager sign-off, IT provisioning"
• Who is involved — "the new hire fills out their details, IT sets up accounts, the manager approves"
• What data to capture — "name, email, start date, equipment checklist, approval checkbox"

The more detail you provide, the better the result. A single sentence like "onboarding process" will produce a generic scaffold. A paragraph describing the specific steps, roles, and data points will produce something much closer to what you need.

Attaching a document

You can optionally attach a file to provide additional context alongside your prompt. Supported formats are PDF, DOCX, Markdown, CSV, and plain text files (up to 5 MB). The document content is included as context for the AI — useful when you have an existing process document, SOP, or checklist that describes the workflow you want to build.

Click "Attach file" below the prompt field to add a document.

Choosing a provider and model

Select which AI provider and model to use from the dropdowns below the prompt. Your selection is remembered for next time.

Generating

Click Generate to submit your prompt. Generation typically takes 5–15 seconds depending on the provider and model.

Reviewing the proposal

After generation, you'll see a preview of the proposed Map showing:

• Title and description — what the AI named your Map
• Stages — each stage with its title, description, assigned role, and list of blocks
• Transitions — how stages connect (shown as a mini graph if the flow is non-linear)
• Roles — the roles the AI created, with icons and colors

Review the proposal to check that the stages, blocks, and flow make sense for your process.

Refining, retrying, or starting over

From the preview, you have several options:

• Refine — describe what you want to change (e.g., "add a compliance review stage after manager approval" or "remove the phone number field"). The AI will adjust the proposal based on your feedback while keeping the rest intact.
• Retry — regenerate with the same prompt and model. Useful if the first result wasn't quite right and you want to see a different interpretation.
• Start over — go back to the prompt screen to write a new prompt from scratch.
• Create Map — accept the proposal and create the Map. This opens it in the builder where you can make further adjustments.

What you get

When you create a Map from a scaffold, Eddy builds the full structure:

Real blocks with data bindings

The AI chooses specific block types based on what each field needs — text fields, date pickers, checkboxes, email inputs, attachment uploads, and more. Each block is created as a real, functional block (not a placeholder) with:

• The correct block type (e.g., Text Field for names, Date for deadlines, Checkbox for approvals)
• A bound column in the Map's data sheet, so responses are captured automatically
• A label and description from the AI's suggestion

Content blocks are the exception — these are static text blocks for instructions or context. They don't capture data and have no column binding.

Pre-populated options

For choice blocks (Single Select, Multi-Select, and Checklist), the AI suggests a set of options. For example, a "Department" single select might come pre-populated with "Engineering", "Design", "Marketing", and "Sales". You can edit, add, or remove options in the builder.

Roles

The AI creates distinct roles based on who is involved in the process (e.g., "New Hire", "IT Admin", "Manager"). Each role is assigned a type, icon, and color. Roles are assigned to stages based on who the AI determines is responsible for that step.

Transitions

If the process is purely sequential (step 1 → step 2 → step 3), stages are connected in order automatically. If the AI detects parallel tracks, alternative paths, or review loops, it creates explicit transitions to represent the flow.

Data sheet

Every scaffolded Map comes with a pre-populated data sheet. The columns in the sheet correspond to the blocks in your Map — so a "Full Name" text field on the first stage creates a "First Stage — Full Name" column. This means your sheet view is ready to use as soon as sessions start.

Tips for good results

• Be specific about roles. "The applicant submits, the hiring manager reviews, HR gives final approval" produces better role assignments than "people do stuff."
• Mention the data you need. "Collect their name, email, phone number, and start date" results in the right block types with proper data bindings.
• Describe the flow. "After the manager reviews, it either goes to HR for final approval or loops back to the applicant for revisions" helps the AI create the right transitions.
• Attach existing documents. If you have an SOP, process document, or checklist, attach it as a file. The AI will use it as context alongside your prompt.
• Use refine for adjustments. Don't try to get it perfect in one go. Generate a first draft, then use the refine option to add or remove stages, change block types, or adjust the flow.

Limitations

• Block options may be generic. The AI suggests options for select and checklist blocks, but they may not match your specific needs. Edit them in the builder.
• No conditional logic. The scaffold doesn't include transition rules or section visibility conditions. Add these in the builder after creation.
• No per-participant stages. All stages are created in shared mode. Switch to per-participant mode in the builder if needed.
• Content blocks are placeholders. Static content blocks contain suggested text, but you'll likely want to rewrite them with your own instructions and formatting.