Skip to content

Welcome to YData SDK Documentation

Pythonversion

Overview

YData SDK is the essential Python toolkit for privacy-first AI and agentic AI development. As AI systems grow more autonomous — reasoning, planning, and acting across complex workflows — the data that trains, evaluates, and grounds them must be high quality, safe, and private by design.

YData SDK closes that gap. It gives teams everything they need to build AI with confidence: synthetic data that mirrors reality without exposing individuals, document and text generation for training and evaluating LLMs and agents, and deep data quality tooling to catch problems before they reach production.

  • Privacy-First Synthetic Data: Generate tabular, time-series, relational, and document data that is structurally and statistically faithful — with no real personal information
  • Agentic AI Training Data: Produce Q&A pairs, synthetic documents, and structured datasets purpose-built for fine-tuning, RAG evaluation, and agent benchmarking
  • Data Quality at Scale: Profile, audit, and improve datasets before they power decisions, models, or autonomous systems
  • Multi-LLM Backend Support: Plug in Workbench, OpenAI, Anthropic, or Gemini — the same API, any provider

Key Benefits

  • Private by Design

    • Synthetic data with zero real PII — compliant with GDPR, HIPAA, and internal data governance policies
    • Safe data sharing across teams, vendors, and regulatory environments
    • PII detection and anonymization built into the same pipeline

  • Built for Agentic AI

    • Generate the training and evaluation data that LLM agents actually need: documents, Q&A pairs, relational schemas, and edge-case scenarios
    • Scanned document simulation for OCR and vision model training
    • Structured error injection (format violations, missing values, referential integrity breaks) to stress-test agents and pipelines

  • Production-Grade Quality

    • State-of-the-art statistical profiling for tabular, time-series, and unstructured data
    • Automated quality scoring and drift monitoring
    • Synthetic data quality reports for model-readiness validation

  • Safe Test & Development Environments

    • Seed dev, staging, and QA environments with realistic synthetic data — no production data ever leaves the secure perimeter
    • Generate relational databases, transaction histories, and document collections on demand
    • Reproduce edge cases, rare events, and failure scenarios that are hard to capture from real data

  • Enterprise Interoperability

    • Native integration with Databricks, Snowflake, and major cloud storage platforms
    • Support for any LLM provider via a unified backend interface
    • Python-native, composable, and CI/CD-friendly

Core Functionality

1. Connectors

2. Metadata

3. Data Profiling

4. Synthetic Data

5. Data Anonymization

Supported Data Formats

Tabular data Synthetic data generator The RegularSynthesizer is perfect for high-dimensional, time-independent data synthesis with exceptional quality results.

Timeseries Synthetic data generator The TimeSeriesSynthesizer handles both regular and irregular time-series data, from smart sensors to stock market data, including support for transactional data with irregular intervals.

Relational databases Synthetic data generator The MultiTableSynthesizer excels at replicating complex relational database schemas while maintaining data integrity and relationships.

The TextSynthesizer and QASynthesizer excels at generating privacy preserving text corpus and generating Question and Answer Pairs for LLM fine tuning and eval.

The DocumentSynthesizer excels at replicating complex custom internal documents while maintaining data consistency and content relevance.