Smart Data Generation

Smart Data Generation is where runs are configured and started. Build a schema, decide how each column is filled, and produce a dataset.

The page has three tabs:

Tab	Purpose
Dataset Generation	Build a full dataset through a guided wizard
Quick Generation	Generate a single value from a reusable template
Dataset Enlargement	Produce more rows that follow an existing dataset's patterns

The active tab is kept in the page URL.

Dataset Generation

The wizard has five steps, plus an optional questions step.

Step	What you do
Data Source	Choose data or add guidance
Questions	Confirm details, when clarification is enabled
Configuration	Set up columns and row count
Review	Review the setup and a generated sample
Generate	Choose output settings and run
Result	Inspect, edit, and download the result

Completed steps can be revisited by selecting them in the stepper. Navigation is locked while a run or an analysis is in progress.

Step 1 — Data Source

Start from data you already have, or from a written description.

• Upload New Dataset accepts .csv, .xlsx, .xls, .json, .db, .sqlite, .sqlite3, and .sql.
• Choose From Uploaded Dataset opens a picker with search and Frequent, Recent, Name, Date, and Rows ordering, and a detail panel showing rows, columns, created date, and creator.
• AI Guidance is a free-text description of the data you need. Analyze guidance turns it into a column plan.

Two switches control how the analysis behaves:

Switch	Default	Effect
Ask questions for ambiguity	On	Adds a Questions step when the guidance is unclear
Data privacy	On	Keeps uploaded values private. Turning it off lets the AI use a small sample from the uploaded data for a closer match

Additional Sources offers two alternative starting points:

• Start from Scratch — define every column by hand, with no source dataset.
• Generate with Service — call an external service and collect its responses instead of synthesizing rows. See Service Generation.

Advanced Configuration exposes an AI model picker, available to users who can update DataCrate settings.

The sidebar shows recent and saved sessions, and the page header shows datasets generated, sessions today, and average generation time.

Step 2 — Configuration

Each column gets a name and a generation method. Use the arrows to reorder columns and the trash icon to remove one. Add Column is available when there is no source dataset — dataset-backed columns come from the file.

Dataset Size (Rows) sets how many rows to produce, from 1 to 5000.

Generation Methods

Method	What it does
AI	Generates values from the column's context and your guidance
Generated Pattern	Produces consistent values from a reusable Quick Data Template
Library	Picks values at random from a Value Library
Distribution Preserving Generation	Creates new values whose patterns resemble the uploaded data
Randomize	Fills rows by shuffling and resampling the source dataset's own values

Which methods are offered depends on what the run is based on:

Situation	Available methods
From scratch, no dataset	AI, Library, Generated Pattern
Backed by an uploaded dataset	AI, Library, Generated Pattern, Distribution Preserving Generation, Randomize
Dataset enlargement	AI, Generated Pattern, Distribution Preserving Generation, Randomize

Each method reveals its own field:

• AI shows an AI Guidance box for that column.
• Library shows a Value Library picker with search, plus Upload new to add a .txt or .json list of values.
• Generated Pattern shows a Quick Data Template picker.
• Distribution Preserving Generation and Randomize show a short explanation instead of a field.

Column data types are inferred, from the AI plan or from the source file, rather than chosen by hand. The inferred type changes what else appears:

• Numeric columns get Min Value and Max Value constraints.
• Email columns get an Email Domain Choice of random uniform, company domain, or custom domain.

If the AI cannot produce a valid plan for every column, the wizard stops and asks you to refine your guidance rather than substituting a fallback.

Step 3 — Review

A summary card shows the source, the columns and their methods, and the row count, alongside a Generated Sample Preview built from a small sample run. If the preview is not ready, Retry Preview replaces the continue button.

Step 4 — Generate

Output Configuration controls what gets written:

Setting	Description
Format	CSV, TSV, Excel (.xlsx), Excel 97-2003 (.xls), JSON, JSON Lines, or Parquet
Output Name	Filename for the artifact. Required
Target Folder	Destination folder in the Data Repository

New Folder creates a destination without leaving the wizard, and a preview line shows the full output path.

Generate Data starts the run. Progress is reported live while it runs.

Step 5 — Result

When the run completes:

• Review & Edit Data opens the generated data in an editor.
• Download File saves the artifact.
• Save these generation details stores the run's configuration so it can be reused later.
• Go to Mock Services, Go to Data Sessions, and Start Another Generation move you on.

Linked Tables

The wizard plans linked tables instead of a single table when the analysis decides the data is relational. That happens in two ways:

• The selected dataset carries a relational schema — it was uploaded as a SQLite database (.db, .sqlite, .sqlite3), a SQL DDL file (.sql), or a JSON file whose top level is a tables list.
• Analyze guidance on a from-scratch run decides your description needs multiple related tables.

Step 2 becomes a plan review where you can inspect each table, its columns, and its Foreign Keys & Attribute Relationships. Add Relationship adds a foreign key; composite keys are entered as comma-separated column names in matching order. Columns that participate in a key or a relationship are marked Linked Key Generation or Linked Attribute Generation rather than being given their own method.

Linked runs always output a linked SQLite database — Format is locked to Linked SQLite database (.sqlite). On the result step, Review Linked Tables, Download Database, and Download Excel replace the usual result actions.

Reusing Generation Details

The Generation Sessions panel on Step 1 lists earlier runs under two filters:

• Recent — the latest dataset generation sessions.
• Saved — sessions whose configuration was explicitly saved.

Saved sessions offer Use these details, which replaces your current selections with that session's settings and drops you on the configuration step. DataCrate confirms first, and verifies that the referenced dataset still exists before overwriting your work.

To save a configuration, use Save these generation details on the result step or in a session's Actions card. Only successfully completed dataset, database, scratch, schema-based, and template runs can be saved — service generation runs cannot.

Service Generation

Generate with Service collects data by calling an external endpoint and storing the responses, rather than synthesizing values.

Service Request

Field	Notes
Protocol	REST or SOAP. gRPC is not supported yet
Method	GET, POST, PUT, PATCH, or DELETE. SOAP is always POST
Endpoint URL	The service to call
Headers and Query Parameters	Key-value pairs sent with each request
Request Body	Payload template
Response Path	Dot-notation path to the repeating list inside the response

SOAP requests add SOAP Action and SOAP Operation.

Per-Row Input

Data Source & Field Binding decides what drives each request:

Source	Behavior
None	Calls the service directly, with no per-row input
Dataset	Sends one request per dataset row
Manual rows (JSON)	Sends one request per row in a JSON array you paste in

With a dataset or manual rows selected, you can pick a saved form schema to define the request fields, or use the source columns directly. Field Mapping matches fields to columns automatically by name, and the Request Body Template builds the payload using {{form.fieldKey}} and {{row.columnName}} placeholders. Generate from mapping writes a starting template for you, and a preview shows the body produced for the first row.

Output and Advanced Settings

Output settings cover the run name, Target Rows (1 to 5000 — rows collected, not requests made), output format, filename, and destination folder.

Advanced settings cover the request timeout, initial and maximum retry delay, maximum consecutive failures, maximum empty batches, how often preview rows are flushed, and a switch to include request details such as the request index and status code in the output.

Generate creates the run and opens its session.

Dataset Enlargement

Enlargement produces more rows that follow an existing dataset's patterns.

1. Choose Dataset — pick an existing dataset, or use Upload Files to bring in a new one.
2. Columns are derived automatically and a method is suggested for each. Columns still using the suggested method are marked as AI suggested.
3. Adjust methods in the column table if you want something different.
4. Set output format, name, and destination folder.
5. Set Target Row Count, from 1 to 5000.
6. Select Enlarge Dataset.

If the analysis cannot prepare every column, enlargement stops and offers Retry AI analysis rather than falling back to a guessed plan.

A running enlargement can be stopped with Cancel. When it finishes, the page opens the session automatically.

Quick Generation

Quick Generation produces a single value at a time from a Quick Data Template — a reusable definition of a data pattern such as an email address, a phone number, an identifier, a credit card number, or an IBAN.

• Search templates, sort by name or creation date, and switch between card and list views. The list loads more as you scroll.
• Generate on a template produces a value, which appears in an editable field with a copy button. Regenerate produces another.
• The ... menu offers View, which shows the template definition read-only, and Delete.

Generate Quick Data Template creates a new template with AI. Enter the Column Name the template is for and describe the pattern under AI Guidance. AI works best with common patterns; the generated template is saved under the column's name and becomes available to every Generated Pattern column.

Related Pages

• Data Sessions — watch runs and inspect their output
• Data Repository — where generated artifacts are stored
• Settings — manage the saved Value Libraries used by Library columns