Context is in early access. Early access features are pre-production software under section 1.4 of
the Cloud Services Agreement and are not for production use. To join the
program, request early access.
total_sales column “revenue,” or which of three customer tables is the one that finance trusts. Context fills that gap. It gives AI models business terms, dataset definitions, and column mappings, so that they can translate a question into an accurate query.
Why Context Matters
Context improves three things:- Accuracy: the model maps your business language to the right columns and datasets.
- Speed: replies come back faster.
- Cost: the model uses fewer tokens to find the data that it needs.
Kinds of Context
Connect AI has two kinds of context: global context, which describes your data, and user context, which describes you. The following table compares them.
Connect AI doesn’t rank one kind of context above the other. Both go to the model together, and the model decides how to use them. Put shared definitions and business rules in global context, and personal details, such as your role, the filters you usually apply, and how you want answers, in user context.
What Global Context Contains
Global context forms a graph of the following entity types:- Global Context: terms and rules that apply to every connection and dataset.
- Connection: context for one data source, which applies to all of its datasets.
- Dataset: field-level details for one table or query.
- Metric: a reusable measure on one or more fields.
- Glossary: business terms and rules, optionally linked to columns. For example, a glossary entry can map Revenue to the
total_salescolumn.
How Models Get Context
How context reaches the model depends on how the model connects to Connect AI.Where Context Comes From
You can add global context in the following ways:- Import an existing semantic model: a dbt semantic layer, a Snowflake semantic view, or a Power BI file.
- Generate context with AI.
- Write context in the built-in editor, starting from a template.
Performance
Asearch_context call typically takes 1.5 to 2 seconds.
Early Access
Context is available through the early access program, which accepts a limited number of participants. To apply, request early access. CData replies within two business days, reviews each request, and then turns on Context for your account.Next Steps
Context Editor
Add, import, and review global context.
User Context
Tell the model about yourself.