AI synthetic data · New

Describe a schema. Get realistic data. In one conversation.

An autonomous AI agent asks clarifying questions, plans, and generates relationally-consistent data into a live database — then iterates in chat, exports to any format, and writes matching documents.

wrapar-agent · new dataset
Generate a SaaS billing schema — customers, subscriptions, invoices.
Got it. A few questions so the data stays realistic:
  • How many customers?
  • Date range for invoices?
  • Currency mix?
500 customers · invoices across 2024 · USD + EUR
Plan 3 tables · FK-linked · realistic distributions
customerplanmrr
Marlowe DiagnosticsScale$1,290
Nordkapp LabsTeam€480
Brightfield HealthEnterprise$4,200
Generated 4,812 rows across 3 tables · relationally consistent
agent.wrapar.ai Describe it. Generate it. Export it.

One hot-swappable model layer. No lock-in.

Anthropic Claude OpenAI OpenRouter DeepSeek Fake — deterministic in CI

From a sentence to a full dataset, conversationally

No fixtures to hand-write, no generator scripts to maintain. You talk to an agent; it plans and builds.

01

Describe

Paste a schema or describe it in plain language — tables, fields, and the relationships between them.

02

Clarify

The agent asks the questions a good engineer would — volumes, ranges, distributions — before it writes a single row.

03

Plan

It orders tables by dependency, resolves foreign keys, and picks realistic distributions for every column.

04

Generate

Rows stream into a live per-tenant database — watch the turns arrive over SSE as the agent works.

05

Iterate

"Add 200 churned accounts." "Skew invoices to Q4." Refine in conversation, with snapshot & revert on every turn.

06

Export

Export to any format — streamed in the background and emailed to you the moment it's ready, with matching documents.

An agent that understands your data model

Conversational data agent

Not a form and not a template. A real agent that plans, calls tools, validates its own output, and course-corrects — the way an engineer would.

Relationally consistent

Foreign keys resolve, cardinalities hold, and distributions look real. Data your app and your tests actually accept.

Multi-vendor model layer

One hot-swappable abstraction over Claude, OpenAI, OpenRouter and DeepSeek. No lock-in, deterministic in CI.

Live streaming turns

Watch the agent think, plan, and write in real time over SSE — optimistic UI, never a frozen spinner.

Any-format export

CSV, JSON, SQL, Parquet, SQLite. Streamed off the request path, emailed on completion, purged on a retention timer.

Matching unstructured documents

Generate invoices, clinical notes, or contracts that reference the exact rows in your structured data — so end-to-end tests have documents that actually line up.

Enterprise-grade, off the request path

01 · Isolated per tenant

Every user's data lives in its own database, sealed behind Fernet-encrypted keys. Cross-tenant reads return a clean 404 — never a leak.

02 · Durable by design

Generation and exports run on Celery workers, not the web request. They survive a redeploy, retry on failure, and email you when done.

03 · Reversible & audited

Every agent tool call is recorded in an append-only log, and each turn snapshots — so a one-click revert always takes you back.

Generate your first dataset in a conversation.

Describe a schema, answer a couple of questions, and export realistic data in minutes.