Access enforcement happens at query time, not in dashboards. When an agent answers a question, every row and column is checked against the user's role. Only the data they're permitted to see is returned - built into the execution, not bolted on.
Query execution pipeline
Row-level security and column masking that activate the instant a query runs - role-scoped, warehouse-native, with zero bypass paths for any downstream tool or dashboard.
Row-level security predicates are injected into every SQL statement at query execution - not enforced in a dashboard layer that users can route around. The warehouse sees only the WHERE clause the role is permitted to satisfy.
Column-level masking policies redact or hash regulated fields - PII, salary bands, margin figures - for every role that lacks explicit read grants. Analysts see the schema; they see only the data their role permits.
Access permissions are built on a typed role hierarchy that flows from the warehouse identity layer into every Quaeris query surface. A role grants exactly what it names - no implicit inheritance, no permissive defaults.
Quaeris writes an immutable audit record for every query: who asked, which role resolved, what RLS predicate was injected, and which columns were masked. Compliance teams get a complete, queryable log - no gaps, no sampling.
Dashboard filters are easy to bypass - users export data and see restricted rows.
Traditional BI tools enforce access in the UI layer, not in the query. If a user can export, they can see everything.
Governed semantic layers require manual access policy definition - one rule per role, per metric, per table.
AI agents have no built-in access controls; they return answers without checking the user's permissions.
Access checks run at query time, not at render time. Every SQL execution is sandboxed to the user's role.
Agents inherit permissions directly from your warehouse (Snowflake, BigQuery, Databricks) - no separate access model to maintain.
Row and column restrictions apply equally to every query path - UI, agent, API, and export.
Unauthorized access attempts are logged with full audit trail: who asked, what they asked, when they asked, and what was blocked.
Fine-grained enforcement at every layer - rows, columns, and the SQL that produces the answer.
Filter entire rows based on user role. A sales analyst sees only their region's data; a CFO sees consolidated numbers across all regions. The agent enforces this - the user can't "see around" it by changing the question.
How row filtering worksRestrict columns by role. Salary data hidden from ops teams; pricing hidden from customer-facing reps. The agent knows which columns the user can access and won't return forbidden fields - even if asked directly.
Masking policiesAccess policies run inside the SQL execution, not after. The database itself returns only authorized rows - Quaeris adds no separate layer, no post-processing cache. The source of truth is the warehouse.
Warehouse-native architectureWhen a user submits a question, the agent runs a six-step pipeline. At each step, role-based access checks ensure data never escapes the user's permission boundary.
| Product | Region | Revenue | Unit Price |
|---|---|---|---|
| Enterprise Suite | East Coast | $1.24M | masked |
| Starter Plan | East Coast | $385K | masked |
| Pro Add-on | East Coast | $210K | masked |
Architecture matters. See where enforcement actually lives in each approach.
| Capability | Quaeris (Governed Agent) | Traditional BI Tool | Ungoverned AI Agent |
|---|---|---|---|
| Row-Level Security | Enforced at query time | UI-layer; bypassable via export | Not enforced |
| Column-Level Masking | Masked before agent sees data | UI-layer; bypassable via export | Not enforced |
| Warehouse-Native | Runs inside warehouse; no data copy | Depends on cloud BI vendor | Usually cached or federated |
| Audit Trail for Queries | Every question, user, SQL, result logged | Partial - dashboard views only | Often no query-level audit |
| Role Inheritance | Uses warehouse roles directly | Depends on connector | Separate access model |
| Access Policy as Code | Defined once, enforced everywhere | Per-tool policies | No policy layer |
| Can Bypass via Export? | No - export respects filters | Yes - exports bypass filters | No enforcement on export |
| Multi-Warehouse Support | Snowflake, BigQuery, Databricks, Redshift, Synapse | Usually single warehouse vendor | Usually single vendor |
Traditional BI includes Power BI, Tableau, Looker. Ungoverned AI includes generic text-to-SQL tools and ChatGPT-connected BI.
Same enforcement architecture, purpose-built for each vertical's access requirements.
Traders see positions; risk managers see exposures by desk; compliance sees audit logs. Row-level access per trader ID, column masking on pricing and P&L. EU regulators require audit trails for every access - Quaeris logs every agent query.
Analysts can query patient cohorts without seeing PII. Identifier masking (names, MRNs) is delegated to warehouse-native controls - Snowflake Dynamic Data Masking, BigQuery Column-Level Security. Researchers access de-identified datasets; clinical teams see full records. All queries are logged.
Regional managers see their region's data; external partners in the same warehouse see only anonymized benchmarks. Product pricing hidden from suppliers. SKU-level access control per partner, enforced at every query without a separate policy layer.
Quaeris reads role definitions directly from your warehouse - Snowflake, BigQuery, Databricks. You don't need to replicate roles in a Quaeris admin panel. If a user is granted read access to a table in Snowflake, Quaeris sees that and enforces it. Role changes sync automatically - no manual reconciliation.
See integrationsBehind the scenes, Quaeris translates a user's role permissions into SQL WHERE and SELECT restrictions. A sales analyst's query becomes scoped to their region and access level. The database executes the restricted query - the warehouse is the enforcement point. You can audit the SQL directly.
Audit trail deep-diveWe'll walk through your warehouse setup, show how role definitions translate to governed queries, and let you ask questions under different user roles. 30 minutes, no sales pitch.
