Browse docs
Browse docs
Browse docs

Documentation

title: Nodes and gateways description: Understand Nanami's current routing gateways and the planning-only fallback transport boundary. ---

Section
Browse docs
Path
/concepts/nodes-agents-gateways

title: Nodes and gateways description: Understand Nanami's current routing gateways and the planning-only fallback transport boundary.

Nanami separates routing-capable gateways from the planning-only fallback transport work that may arrive later.

Node

A node is a data-plane participant in one or more workspace networks.

Node records include identity material (public key), status, endpoint metadata, and network memberships.

Gateway workers (current)

A Gateway is a shared platform dataplane worker that can route traffic for assigned workspace networks when the runtime can isolate each tenant/network route domain safely.

Gateway behavior in MVP:

  • runtime inventory is platform/operator managed; workspaces do not own one gateway each,
  • assignment rows, network membership, policy, and audit stay workspace-scoped,
  • shared multi-tenant assignment is product-owned and must fail closed unless route-domain isolation, per-network protocol interfaces, and packet-time source+destination ACLs are available,
  • can participate in ingress/egress and exit-node routing,
  • receives desired state from gateway-manager,
  • applies interfaces/routes/NAT through gateway-daemon,
  • can be assigned automatically (auto, balanced) and exposed through explicit operator controls where those controls are implemented.

Diagram (text)

Client -> Gateway -> Private network resources

Gateway manager and gateway daemon agents

Nanami currently uses two runtime agents:

  • gateway-manager (compat: gateway-manager): desired/observed control loop.
  • gateway-daemon (compat: gateway-daemon): local runtime apply and telemetry.

Diagram (text)

Control plane -> gateway-manager -> gateway-daemon -> observed state -> control plane

Shared region groundwork (current)

The current release already carries shared-region groundwork and gateway-worker inventory that future fallback transport can build on:

  • a platform-scoped region catalog,
  • tenant/workspace access to named regions,
  • internal read-only oversight surfaces for platform operators.

This groundwork is not a live DERP-like fallback transport mesh. It is the current placement and explanation contract for gateway runtime without pretending fallback relay behavior exists today.

Future fallback transport (planning-only)

Fallback transport remains planned as connectivity fallback only (DERP-like), not as a routing gateway or policy router.

Planned properties:

  • stateless forwarding layer,
  • safe to share across workspaces because no route ownership,
  • no workspace AllowedIPs/route/NAT control,
  • explicit separation from the current routing-gateway baseline.

Diagram (text)

Client A <-> Shared fallback transport (future) <-> Client B

Why this split matters

  • Gateways route and can carry workspace-sensitive policy, so tenant isolation remains in assignments and generated runtime state.
  • Shared-region groundwork is current, but shared fallback transport is not.
  • Future fallback transport should improve reachability without becoming a policy router.
  • This keeps cross-workspace safety and operator mental model clear.

Planned follow-up areas

  • Shared fallback transport implementation on top of the current shared-region groundwork.
  • Multi-hop routing chains and policy-based pathing.
  • Gateway redundancy policies (redundancy: 2+).