One engine for SQL, models, AI and streams.
Standard SQL on Apache Iceberg.
Iceberg tables
Create, load, query and evolve Iceberg tables. Snapshots and time travel come from the table format itself.
Transactions and procedures
Multi-statement transactions, stored procedures and control flow for real workflows, with rollback when a step fails.
Load your files
Upload files in Studio or bulk-load with COPY INTO, then query them like any other table.
Connect your tools
Arrow Flight SQL, the PostgreSQL wire protocol, JDBC and a Python SDK. Contact us to enable client access for your organization.
Train and score models without leaving SQL.
CREATE MODEL trains on a query. The model becomes a function you call in any SELECT, so
scoring happens where the data already is.
Native models
Linear and logistic regression, multinomial classification, decision trees, neural networks, K-means clustering, PCA and ARIMA forecasting.
Tuning and evaluation
Search hyperparameters with ML.TUNE, evaluate on held-out data, and version every model you train.
Bring your own model
Import ONNX models, or register a remote model that calls your own HTTPS endpoint.
Features and learning
Feature groups with backfill, and online learning that keeps a model current as new rows arrive.
Language models and retrieval, governed per organization.
AI functions
Classify, extract, summarize, translate, mask and complete text in SQL, row by row or in batches.
Vector and text search
Embed text, build vector indexes, and combine them with BM25 keyword search for hybrid retrieval and grounded answers.
Ask in plain language
ASK answers a question from your tables; SUGGEST QUERIES proposes SQL you can inspect, and conversations keep context across questions.
AI Governance
AI is off until an administrator sets a monthly token budget and approves a policy: provider, model, operations, and data egress and retention.
Fresh data without a separate pipeline.
Stream ingestion
Write events continuously into Iceberg tables, and query them like any other table.
Scoring on arrival
Derived streams score each new row with a model as it lands, for example to flag anomalies.
Dynamic tables and tasks
Keep derived tables within a freshness target, and schedule SQL to run on its own.
The workspace for all of it.
Worksheets
A multi-statement SQL editor with per-statement results, query history, saved and shared queries.
Notebooks
Python notebooks that run in isolated sandboxes next to your data.
Dashboards
Build dashboards from SQL, and use AI to plan, question and explain them when your organization allows it.
Tutorials
Seventeen guided tutorials that run end to end in a new organization.