Context layer security and governance
The context layer enforces security on every question and provides governance controls so you stay in charge of what the Spotter agent knows.
Security
Security filters what reaches the Spotter agent. The checks run on every question, for every context entry, against live permissions, before the prompt is built and before anything is ranked.
Access boundaries
Only context that the user can reach is considered. This includes organization, cluster, data model, agent, user, and group boundaries. Context from a different organization, a different cluster, or an agent or data model the user does not have access to is never retrieved.
Structured data rules
When SQL queries are executed against structured data, the query honors the data model’s permissions, including column-level security (CLS) and row-level security (RLS), for the user who is asking. However, when memory and context are fetched for reasoning, these data model permissions are not enforced. The reasoning output may reference columns and values that are restricted by CLS or RLS rules.
For more information about CLS and RLS, refer to Data security.
Governance
Governance keeps you in control of what the context layer knows and how it evolves.
- Transparency and human in the loop
-
Review what the layer is about to learn before it takes effect. Important changes wait for a steward’s approval.
- Manage
-
Update, delete, and link context entries manually.
- Migration
-
Publish and manage context across agents, organizations, and clusters. Move context, reassign it, or take it with you to a new environment.