Shared knowledge. Defined access.

The identity and access controls available in Palace Server.

Organization boundaries

PostgreSQL row-level security separates organization data. The starter kit uses a dedicated non-superuser database role so those boundaries are enforced. Preserve that database-role setup when adapting the deployment to your infrastructure.

People and agents

Palace has six roles: owner, admin, maintainer, developer, read-only agent, and security auditor. Choose roles according to the work each identity needs to perform. Coding tools authenticate with per-user API tokens that are stored hashed with Argon2id and can expire or be revoked. New keys default to the developer role and a 60-day expiry.

Admins add people with OIDC sign-in or, from v0.8.0, with the admin API, which creates the user, takes one seat, and issues their first token. See Add your team. Each memory records who filed it, and palace_audit_explain shows its history.

Single sign-on uses OIDC with the authorization-code flow and PKCE. Identity-token signatures are verified against the provider’s published keys.

What the audit log covers

Memory writes have tenant-scoped audit records that name the acting user, and MCP tool calls create usage events. From v0.8.0, adding a user and creating or revoking an API key are audited too. Plan your wider operational logging, such as sign-ins, around that scope.

  • Keep individual API credentials separate from the server license.
  • Restrict database and admin endpoints to trusted networks.
  • Terminate TLS at your reverse proxy and verify the database TLS connection.
  • Restrict the /metrics endpoint at your proxy.
The IT evaluation brief describes these controls for the current release on one page.