Skip to main content

Data handling Private Preview

This page covers where your data goes during each stage of Context Engineering Studio. Use it for internal security reviews and governance assessments.

Data flows​

Each row below maps a CES operation to where the data travels and where it stays.

FlowWhere your data goes
Asset discoveryAtlan catalog only. No warehouse query runs.
Column metadata fetchWarehouse INFORMATION_SCHEMA → Atlan backend → your browser.
AI model generationCatalog metadata → Atlan AI → structured suggestions back to your browser. Atlan AI generates synthetic examples rather than consuming real warehouse rows.
Snowflake chat and simulationQuestion + YAML → Cortex Analyst (in-account) → SQL + results. Judge runs in Snowflake via SNOWFLAKE.CORTEX.COMPLETE(). Result data stays in Snowflake.
Databricks chatQuestion → Genie Space (in-workspace) → SQL + results. Result data stays in Databricks.
Databricks simulation judgeNatural-language question + generated SQL + verified SQL → Atlan AI → semantic-equivalence verdict.
DeploymentDDL + YAML → Snowflake (in-account via SYSTEM$CREATE_SEMANTIC_VIEW_FROM_YAML) or Databricks (in-workspace via CREATE OR REPLACE VIEW ... WITH METRICS LANGUAGE YAML plus Genie Space API calls).

Data residency​

Where data resides and how it moves at each stage, per engine.

  • Chat and Simulate: question text + YAML travels to Cortex Analyst, which runs inside your Snowflake account. SQL generation, execution, and the simulation judge all happen in-account. No query result data leaves Snowflake.
  • Deployment: CES calls SYSTEM$CREATE_SEMANTIC_VIEW_FROM_YAML in-account. The deployed Semantic View lives in the database and schema you chose.
  • Observability: CES reads production traces from SNOWFLAKE.LOCAL.CORTEX_ANALYST_REQUESTS_V, a built-in system view, and syncs them into the CES Observe tab.

Authentication​

CES authenticates to each engine using the credentials configured on your existing Atlan connection.

  • Snowflake: Atlan authenticates via the service account role on your existing Snowflake connection (Personal Access Token or RSA keypair).
  • Databricks: Atlan authenticates via Service Principal + OAuth M2M on your existing Databricks connection. PAT is supported as a fallback but isn't recommended for production.

See Grant Snowflake permissions and Grant Databricks permissions for setup detail.

Atlan AI security model​

Atlan AI operates under a metadata-only security model:

  • Atlan AI doesn't have direct query access to your warehouse and can't issue arbitrary queries on your tables.
  • Atlan AI operates on metadata (schemas, column names and types, catalog descriptions, glossary terms, lineage, and query-history patterns), not row data.
  • Where sample examples are needed, Atlan AI generates synthetic examples based on sample-value characteristics already captured in the catalog. Actual rows from your source tables aren't sent to Atlan AI.

Security review reference​

Common vendor security and governance questions, mapped to CES-specific answers.

ConcernDetail
Atlan AI access scopeMetadata-only. Atlan AI has no direct query access to your warehouse and doesn't consume row data. Sample examples are synthesized from catalog metadata.
Result data during simulation judgingSnowflake keeps result data in-account via SNOWFLAKE.CORTEX.COMPLETE(). On Databricks, Atlan AI receives only the natural-language question and both SQL statements. No warehouse row data is sent.
Data stored by AtlanCES persists session state only (selected assets, question sets, evaluation results, current YAML), in object storage managed by Atlan's infrastructure layer. No warehouse query results are stored.
Deployed artifactsAll deployed Semantic Views (Snowflake), Metric Views, and Genie Spaces (Databricks) live inside your account, under your governance. Atlan holds no copy.

For a signed data-handling attestation, contact Atlan support.

See also​