Dataverse Sample Data Generator

The first tool that generates contextually meaningful sample data for Microsoft Dataverse — powered by AI.

The Problem

Every Dataverse consultant, developer, and tester faces the same challenge: generating realistic sample data is painful.

• Building a demo for a client? You need 200 accounts, 500 contacts, and 1,000 cases — and they all need to look real.
• Setting up a trial instance? You're manually typing records one by one, or importing flat CSVs with no relationships.
• Testing a Power Automate flow? You need data across multiple related tables, with proper lookups wired up.

Existing tools can generate random data — random names, random numbers, random strings. But no tool in the market generates data that is contextually meaningful. You get "John Doe" and "123 Main St" regardless of whether you're building a healthcare app in Singapore or a retail chain in Germany.

Until now.

What Makes This Different

This extension uses the AI language models built into GitHub Copilot to generate data that actually makes sense for your scenario. It doesn't just fill columns with random values — it understands your business context and produces data that tells a coherent story.

Business Context Changes Everything

Give the tool a simple description like "Healthcare clinic in Singapore" or "Law firm in London", and the AI generates data that fits — region-appropriate names, industry-specific job titles, realistic case descriptions, matching email domains. Across related tables, the context stays consistent: accounts, contacts, and cases all tell the same coherent story.

Without context, you get Test Account 47 and City_0293. With context, your demo environment looks like production.

Why a VS Code Extension?

This tool is built as a VS Code extension by design — not a web app, not a standalone tool. Here's why:

It runs on GitHub Copilot's AI models. The contextual data generation is powered by the language models available through GitHub Copilot Chat. This means:

• You need VS Code with GitHub Copilot installed
• The AI generation uses the same models that power Copilot's code suggestions
• No API keys to manage, no separate AI subscriptions, no token costs — if you have Copilot, you have AI data generation
• When the LLM is unavailable, the tool automatically falls back to Faker.js so it always works

This tight integration with Copilot is what enables the contextual generation that no other tool offers.

Who Is This For?

• Consultants building demo environments for client presentations — stop spending hours manually creating realistic-looking data
• Developers who need test data across related tables — generate 500 records across 10 tables in one click instead of writing import scripts
• Testers validating business logic — get data that actually exercises your workflows, not random noise
• Solution architects setting up trial instances — populate an entire Dynamics 365 environment in minutes, not days
• Trainers preparing training environments — generate consistent, realistic data that trainees can relate to

Save Hours, Not Minutes

Populating 5 related tables with 100 records each — complete with proper lookups and realistic values — typically takes 2-4 hours of CSV prep, manual imports, and relationship fixing. With this extension, it takes about 2 minutes. Describe your business context, pick your tables, click Generate. Need to tear it down and start fresh? Cleanup mode handles that in seconds, respecting referential integrity automatically.

Features

Generate Mode

• AI-powered contextual generation — describe your business scenario and get data that fits
• Document upload for context extraction — upload a requirements document (.docx, .pdf, or .txt) and the AI automatically extracts business context and auto-selects matching Dataverse tables. No manual typing needed — just upload your spec and generate.
• Multi-table support with automatic dependency resolution (topological sort)
• Relationship handling — automatically links records via lookups using @odata.bind
• Column filtering — Best Applicable, Only Mandatory, or hand-pick specific columns
• Faker.js fallback — always works, even without Copilot
• Batch writes using Dataverse $batch API for efficient bulk inserts

Cleanup Mode

• Safe deletion with reverse topological sort (children deleted before parents)
• Sort order — delete newest or oldest records first
• FetchXML filter — target specific records with custom FetchXML queries
• Deletion plan preview — review before executing irreversible deletes

GitHub Copilot Chat Integration

• Chat Participant (@dvdata) — generate data through natural language
• Language Model Tool — other Copilot extensions can invoke data generation
• Commands: /connect, /tables, /plan, /generate, /cleanup

Authentication

• Browser Sign-in (recommended) — supports MFA, SSO, federation via PKCE
• Device Code flow — for restricted environments

Getting Started

Prerequisites

• VS Code 1.100.0 or later
• GitHub Copilot extension (required for AI-powered contextual generation)
• A Microsoft Dataverse environment (Power Platform / Dynamics 365)
• Azure AD account with read/write access to the target environment

How to Use

1. Install the extension from VS Code Marketplace
2. Open the Command Palette (Ctrl+Shift+P on Windows / Cmd+Shift+P on Mac)
3. Search for and select "Dataverse Sample Data: Open Generator UI"
4. The Generator UI opens as a panel inside VS Code — everything happens here:
   • Connect — enter your Dataverse environment URL and sign in via browser
   • Set Context — type a business description, or upload a document (.docx, .pdf, .txt) to auto-extract context and auto-select matching tables
   • Select Tables — pick the tables you want to populate (or let document upload auto-select them for you)
   • Configure — set record count, column mode
   • Generate — click and watch real-time progress as AI-generated records are created

Tip: You can also access the tool through GitHub Copilot Chat using @dvdata — see the Chat Integration section below.

Using with Copilot Chat

@dvdata /connect https://yourorg.crm.dynamics.com
@dvdata /generate account, contact — 50 records each for a healthcare clinic in Singapore
@dvdata /cleanup account — delete the 50 newest records

How It Works

┌─────────────┐    ┌──────────────┐    ┌────────────────┐    ┌──────────────┐
│  Connect &  │───▶│  Read Table  │───▶│  AI generates  │───▶│  Batch write │
│  Authenticate│    │  Metadata    │    │  contextual    │    │  via $batch  │
│  (MSAL/PKCE)│    │  & Relations │    │  data per row  │    │  with lookups│
└─────────────┘    └──────────────┘    └────────────────┘    └──────────────┘
                                              │
                                    ┌─────────┴──────────┐
                                    │  Fallback: Faker.js │
                                    │  (if LLM unavailable)│
                                    └─────────────────────┘

1. Metadata Discovery — reads table schemas, column types, and relationships from Dataverse
2. Dependency Planning — topological sort ensures parent records are created before children
3. AI Generation — sends column metadata + your business context to GitHub Copilot's LLM, which returns realistic values
4. Batch Writing — creates records using Dataverse $batch API with @odata.bind for lookups
5. Result Reporting — shows per-table success/error counts and data source (AI vs Faker)

Extension Settings

Known Limitations

• Maximum 5,000 records per generation run (configurable)
• FetchXML cleanup queries limited to first page of results
• Some system columns are auto-excluded (state, status, auto-number, computed)
• AI generation requires GitHub Copilot — falls back to Faker.js without it

License

MIT