Security model
Understand Maia's identity gates, authorization layers, approvals, and audit boundary.
Maia's security model layers platform identity, governance authorization, tool exposure, file and terminal policy, human approvals, and audit evidence. No single layer replaces the others.
Security layers
- Platform authentication identifies the sender.
- Gateway admission determines whether that identity reaches the gateway.
- Governance mapping assigns roles and teams.
- Tool configuration limits the actions exposed on that platform.
- File and terminal policy authorizes concrete resources and operations.
- Human approval decides protected changes or runs.
- Audit logging records sensitive decisions and mutations.
Rely on Maia's fixed default deny
Governance is always enabled and cannot be turned off. An unmatched file path is always denied, and a role alone grants no file access. Gateway shell and code execution are isolated to explicit grants through Docker or Podman; a missing secure-runtime boundary causes a denial, never host fallback. Maia calls this safe degraded state Restricted mode.
Add narrow person, team, role, or folder-policy grants for defined work. Avoid broad home-directory or filesystem-root permissions.
Separate user and administration surfaces
Most users interact through a messaging gateway. The dashboard can manage secrets, providers, policies, sessions, jobs, plugins, and approvals, so dashboard access must remain narrow and authenticated.
This Ampliia site is public product documentation. It neither discovers nor connects to a Maia installation. Dashboard addresses, credentials, and tokens remain inside the deployment.
Treat approvals as decisions
An approval must identify the requester, target, operation, proposed content or command, approver authority, and result. Do not create automatic approval rules that reproduce the permission you intended the checkpoint to constrain.
Protect the audit trail
- Enable audit logging before broad rollout.
- Redact sensitive values.
- Set retention according to company and regulatory needs.
- Restrict write access to audit storage.
- Export events to a protected external system when independent retention is required.
Know the boundary
Governance reduces risk but does not make unsafe models, tools, plugins, MCP servers, host permissions, or network exposure safe. Apply operating-system isolation, network controls, credential scoping, backups, monitoring, and human review appropriate to the deployment.
License and attribution
Maia is an Ampliia distribution derived from upstream Hermes Agent components. Review the repository license and bundled third-party notices before redistribution or commercial deployment.