Security Model

Security is foundational, not a feature bolted on. The guiding rule: everything authenticates, everything authorizes, everything is audited, nothing bypasses security.

Principles#

• Authenticate everything — humans via SSO, services via keys/mTLS, agents via verifiable credentials.
• Authorize every action — server-side, against capability, tenant and (soon) policy.
• Audit everything — every meaningful action emits an immutable event.
• Least privilege — scope tokens, agents and MCP access to the minimum required.

Secrets are never embedded#

Agents should not carry long-lived secrets in their configuration. AuthSpoke is designed to store encrypted secrets centrally and issue short-lived, scoped credentials — with future integrations for HashiCorp Vault and the major cloud key vaults. Until managed secrets ship, store credentials in a vault and inject them at runtime.

Operational safety#

Kill Switches let operators contain an incident in seconds — disabling agent execution, MCP traffic, external LLM calls or token issuance with no redeploy, and an audit record of every change.

Tenant isolation#

Every query, event and secret belongs to a tenant, enforced server-side. See Multi-Tenancy.