Introduction to Semantic Data Models
Semantic Data Models is currently an experimental feature, available starting from Studio 1.6.3-preview. You will need to enable the Experiments in the Advanced Settings to use it. While the feature is in preview, it is not available in Agent Compute for deployed agents yet.
What it does
The Semantic Data Model is the agent’s context layer for enterprise data. It links agents to the exact tables, views, and files an agent can use, layers on business meaning, and turns natural-language prompts into engine-specific SQL. Results flow back as DataFrames inside the Agent Server, so follow-up questions stay grounded in the same dataset without round-tripping data through the LLM.
Why teams use it
- Connect agent to databases (Postgres, Snowflake, Redshift, and more coming soon).
- Train agent on how to understand and use spreadsheet uploads without writing code.
- Pick the most relevant tables/views and columns and keep business context in sync at table and column level.
- Let the agent generate dialect-specific SQL, reuse Verified Queries for recurring workflows, and keep every answer consistent with policy.
- Work with millions of rows directly inside DataFrames, enabling iterative filtering, joining, summarizing, and charting in chat. And most important - without blowing up the context window and token costs.
Key concepts
Data Connections – Secure connections configured in Studio or Control Room to connect to your data sources.
Chat Files – Uploaded files that are available in the chat, and can be used with a Semantic Data Model to help agent understand the data and answer questions.
Business Context – Freeform text that describes the data from the business perspective. This is used to help Studio generate the Semantic Data Model automatically.
Verified Queries – Saved SQL queries (for example, Smooth Timeseries Data or Create Timescale Projection) that agents can reuse trusted without going through the NL2SQL generation every time.
DataFrames – DataFrames are the agent's workspace for query responses. They are created from the your query results and can be further transformed and analyzed using the agent's natural language capabilities.
Start building Semantic Data Model
Start building a Semantic Data Model in Studio from any agent, open it's information side panel from the button in the top right corner. You will see the Data tab in the side panel.
Once you click the button, you will see options on what type of data you want to connect to your agent, or for importing an existing Semantic Data Model.
Follow the instructions for the type of data you want to connect to your agent below: