Dex Semantics
Dex is an agent-native analytics engine for exploring, transforming, and maintaining data warehouse analytics with read-only access, local context, and scan cost controls.
Dex is an agent-native analytics engine for exploring, transforming, and maintaining data warehouse analytics with read-only access, local context, and scan cost controls.
What the product does and how it is positioned
Dex functions as an analytics interface for AI coding agents, facilitating data exploration, model construction, and ongoing system maintenance. It utilizes a read-only architecture to profile data sources, flag personally identifiable information, and cache structures locally, so that raw data and credentials remain separate from the LLM context.
The system includes guardrails designed to manage resource consumption by enforcing dry-run estimations for data scan volume before operations run. It provides tools for defining semantic layers and models, while offering maintenance capabilities to monitor for schema, grain, and semantic changes over time.
Source-supported ways to use the product
AI agents utilize exploration tools to profile columns, infer table relationships, and generate a draft map of the data warehouse.
Developers use the system to generate dependency-aware SQL models and semantic layer definitions compatible with frameworks like dbt and Cube.
The documented workflow, where available
The agent identifies and profiles data sources to build a local map, flagging PII and inferring table relationships.
The agent generates models and semantic definitions based on specific data requirements while validating against dev targets.
The system checks established data snapshots to detect and propose fixes for schema, grain, and semantic drift.
The system is designed to interact with data warehouses without compromising the underlying security of the raw data. By operating with a strictly read-only approach, the engine is designed so that raw rows and specific sensitive values are not copied out of the source environment.
Security is maintained by keeping credentials local to the user's agent and so that they are never included in the context window sent to the LLM. Only bounded summaries, schema information, and statistics are cached locally to facilitate agentic understanding.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
Dex connects to several major data warehouses including BigQuery, Snowflake, Databricks, Redshift, ClickHouse, PostgreSQL, Trino, and DuckDB.
Dex operates with read-only access, storing only bounded summaries and aggregates locally, while keeping credentials and raw data separate from LLM contexts.
Yes, the maintenance module detects schema, grain, and semantic drift by comparing current data against previously established project snapshots.
Dex enforces cost guards by requiring dry-run estimations for data scans before allowing the agent to execute queries or transformations.
Dex supports authoring for dbt Semantic Layer, Apache Ossie, SLayer, Cube, Snowflake Semantic Views, and Databricks Metric Views.