Security and permissions for company context.
Updated 2026-06-07 · Agent-readable markdown available
The controls required before company context can safely power AI agents.
Company context becomes risky when it lacks source, scope, redaction, revocation, conflict review, and audit. Ambience makes those controls part of the memory layer.
- Agents should not receive context only because it is relevant.
- Raw secrets should not become durable memory.
- Admins need proof of who read which context and why.
Relevance is not permission
A memory can match a query and still be inappropriate for the current user, project, or agent. Company context needs access control before retrieval reaches the prompt.
The trust controls
Ambience uses redaction before storage, five scopes, source-linked provenance, revocation, conflict review, and append-only audit.