Security & Governance
How the Company Brain governs access to shared knowledge: permission-scoped access, central policy, authentication, data removal, and grounded answers for enterprise review.
The Company Brain is a shared knowledge layer, so governance and access control are foundational, not add-ons. This page describes the controls that determine which people and agents can reach what.
Governed by company policy#
Knowledge and memory are company-shared, but access is never wide-open. It is governed by your company's permissions and roles, enforced centrally for every agent and every source:
Every agent reaches only the knowledge and memory it has been granted. Permissions are defined by your company and applied to every request.
Access rules are defined once and enforced everywhere for every agent, application, and connected source, rather than re-implemented per integration.
For stronger separation, the platform also supports isolated single-tenant and sandboxed / self-hosted deployments where data stays within your boundary. See Deployment Models.
Authentication#
Every request to the Company Brain is authenticated with your company credentials before it touches any data. Credentials carry the access an agent is entitled to, so the governance layer can scope each request to exactly the knowledge it's permitted to see.
Credentials are managed centrally and should be kept server-side, never embedded in client-side or browser code. Access can be granted, scoped, and revoked from one place. That same place records every grant for audit.
Data lifecycle & removal#
- Kept current. As connected sources change, the brain re-syncs so knowledge stays accurate.
- Permanent removal. Content can be permanently removed from the brain by source, by tag, or in bulk, giving teams a clean primitive for honoring data-removal and retention requirements.
- Reversible refinements. Changes to the meaning of your data (the semantic layer) are reversible, so corrections never risk the underlying facts. See the Ontology Graph.
Grounding as a safety property#
Because answers are derived from your real data and returned with citations and a confidence signal, agents and applications can verify every answer against its sources rather than trusting free-form generation. The confidence signal lets an application decide how prominently, or whether, to surface an answer.
Identity, access governance, and auditability.