Skip to main content

What is PostgresML?

PostgresML is an end-to-end machine learning platform built entirely inside PostgreSQL. It allows you to train models and make predictions using SQL queries, without moving data out of your database.

Overview​

PostgresML transforms your PostgreSQL database into a complete machine learning platform by adding support for training and deploying models directly within the database. This eliminates the need for external ML infrastructure and data pipelines.

Key Features​

🤖 In-Database ML

  • Train models directly on your data without ETL
  • Use familiar SQL syntax for machine learning
  • Real-time predictions with low latency
  • No data movement reduces security risks

🎯 Pre-trained Models

  • Access to Hugging Face transformers
  • State-of-the-art NLP models
  • Computer vision models
  • Time series forecasting models

⚡ Vector Operations

  • Built-in vector similarity search
  • Store and query embeddings efficiently
  • Semantic search capabilities
  • RAG (Retrieval Augmented Generation) support

🔧 Multiple Algorithms

  • Classification and regression
  • Clustering and dimensionality reduction
  • Natural language processing
  • Time series analysis

How PostgresML Works​

PostgresML extends PostgreSQL with machine learning capabilities through a native extension. The workflow is simple:

  1. Store your data in PostgreSQL tables
  2. Train models using SQL functions
  3. Make predictions with SQL queries
  4. Evaluate and deploy models in production

This approach provides several advantages:

  • No data movement between systems
  • Leverage PostgreSQL's ACID guarantees
  • Use existing database security and permissions
  • Scale with your database infrastructure

Use Cases​

Real-Time Recommendations​

  • Product recommendations in e-commerce
  • Content personalization
  • User behavior prediction

Natural Language Processing​

  • Text classification and sentiment analysis
  • Named entity recognition
  • Question answering systems
  • Semantic search

Fraud Detection​

  • Real-time transaction analysis
  • Anomaly detection
  • Risk scoring

Time Series Forecasting​

  • Sales forecasting
  • Demand prediction
  • Resource planning

Vector Search & RAG​

  • Semantic document search
  • Chatbots with context
  • Knowledge base queries

Architecture Components​

PostgresML Extension​

A PostgreSQL extension written in Rust that provides machine learning functions and operations directly in the database.

Dashboard​

A web application for managing models, monitoring performance, and visualizing results.

SDK​

Client libraries for Python, JavaScript, and other languages to interact with PostgresML programmatically.

Model Store​

Built-in model registry for versioning and deploying trained models.

Comparison with Other Solutions​

FeaturePostgresMLExternal ML ServicesStandalone ML Platforms
Data MovementNoneRequiredRequired
LatencyVery LowMedium-HighMedium
Setup ComplexityLowMediumHigh
SQL IntegrationNativeVia APIsVia ETL
CostDatabase onlyAdditional servicesAdditional infrastructure

Supported Algorithms​

PostgresML supports a wide range of algorithms through scikit-learn, XGBoost, LightGBM, and more:

  • Linear Models: Linear/Logistic Regression, Ridge, Lasso
  • Tree-based: Random Forest, Gradient Boosting, XGBoost, LightGBM
  • Neural Networks: MLP, Deep Learning via transformers
  • Clustering: K-Means, DBSCAN, Hierarchical
  • Dimensionality Reduction: PCA, t-SNE, UMAP
  • NLP: Transformers from Hugging Face

Getting Started​

Ready to try PostgresML? Continue with the Installation Guide to set up PostgresML on your PostgreSQL database.

Resources​