One control plane. Explicit responsibilities.
Access administration and traffic handling have different jobs. The control plane manages organizations, members, access services, and access lifecycles. Nodes execute that intent on infrastructure controlled by the operator. Customer traffic passes through the nodes, never through the control-plane process.
Organizations are a data boundary
Organization-owned data is stored in a separate database for each organization. Authentication and trusted organization resolution select the organization before its application logic runs. Global identity, sessions, and organization routing live in the control-plane database.
Each node belongs to one organization. That organization owns its services, member-owned access, connection records, and operational data. This is application and database separation; it is not a claim of separate physical hosts or protection from a compromised control-plane host.
Access has an owner and a lifecycle
A member can own multiple Accesses. Each provisioned Access has one current Connection, served by an access service on a node. Suspension is reversible; revocation is terminal and preserves historical records.
Members explicitly retrieve their own connection configuration. Administrators manage the lifecycle and see issuance results, but cannot retrieve another member’s existing secrets through ordinary administration workflows.
Nodes pull intent and report application
The node agent initiates management communication. It polls for desired state, verifies the Ed25519 signature, reconciles local configuration, and reports its applied revision and health. No inbound node management port is required. The control plane does not need SSH or root credentials for the node host.
The current runtime adapter applies a combined configuration snapshot. Local configuration validation, atomic replacement, reload, and health checks guard that process. Failed application triggers recovery of the previous configuration where possible; recovery status must still be checked.
A saved revision is not proof of application. An applied revision is not a test of external connectivity. Keeping these observations separate helps administrators diagnose changes without overstating their result.
Runtime integration belongs on the node
WireGuard and Shadowsocks are currently implemented through sing-box. The control plane describes protocol intent; the agent owns integration with the networking engine. Native WireGuard, shadowsocks-rust, general engine discovery, and advanced transports are future work.
Availability has a tradeoff
Nodes retain their last known good configuration during control-plane downtime. This lets existing local configuration continue, but an offline node can also retain an earlier authorization. Immediate revocation on disconnected nodes and local offline expiry are not implemented. Check node application status after changing access.
Collect operational signals, not packet contents
The control plane may store identity, access, audit, health, version, and coarse byte-counter data required to administer the service. Normal operation does not require packet contents, browsing history, DNS queries, or destination history. This is a description of the current data model, not a certification or a blanket anonymity guarantee.
Choose where the control plane runs
The same architecture supports a hosted control plane with remote nodes, a customer-operated control plane with remote nodes, or a customer-operated single VPS with a colocated node. Availability of a hosted service and commercial support require a separate offer. The guided single-VPS workflow is currently alpha.
Go deeper
Review the current compatibility and security documentation before deploying.
Documentation access