Communication and storagePeer transports and discovery

Peer transports and discovery

Find peers, select permitted paths, and separate Atlas protocols from external interoperability.

Use discovery to find candidate peers and transport to establish a permitted path. Authenticate the peer identity, then authorize the operation. Discovery and connectivity alone grant no access.

An IP-capable network can carry native QUIC peer communication. The network can be Wi-Fi, cellular, Ethernet, or a radio that exposes an IP interface. The radio's brand does not determine the Atlas protocol.

A non-IP link needs a transport adapter that implements framing and the required Atlas handshake. A constrained message-only link does not become a bidirectional byte stream because another target supports one.

Browser WebAssembly uses its browser adapter and WebSocket constraints. It does not expose raw QUIC, native radio access, or native discovery merely because another SDK target does.

Declare protocols and exchange intent

A Protocol ID names a versioned exchange such as fieldnote/telemetry/1. The format is {namespace}/{name}/{version} with a positive integer version. atlas/ and atlas-mq/ are reserved namespaces. Use your own namespace.

Atlas exchanges a set of supported Protocol IDs. This differs from TLS ALPN, which selects one application protocol for a connection. Do not call the Atlas capability set ALPN.

Declare an exchange intent with the target peer, exchange kind, payload hint, transport constraints, and priority. The transport manager schedules connections. Keep connection and retry work outside UI startup.

A TransportStream is a handler's bidirectional byte interface, not the transport connection or authenticated transport session. These have different lifetimes. Stream handlers require stream-capable transports.

Introduce and resolve peers

Start with a trusted introduction that binds an expected identity to an address or discovery channel. Advertise only the application metadata you intend to disclose. Observe discovery changes, then resolve current addresses when starting an exchange.

A discovered peer can be stale, unreachable, or unauthorized. Treat discovery as a hint, not an admission record. Verify the authenticated identity against the intended peer before exchanging application data.

An application's preferred peer flag can guide ranking. It does not guarantee reachability or confer a special Platform role. An always-on application peer, an infrastructure transport relay, and an external protocol adapter solve different problems.

Atlas Gateway Fabric's WebSocket and MQTT carriers remain domain-neutral. The deployed Gateway Fabric process also runs a persistent Mission operational peer. Reusing a carrier does not require Mission admission. Independent Platform apps can communicate directly or run their own always-on peers without Gateway Fabric.

Route without changing the contract

When a permitted direct peer path exists, events and live samples can use it without waiting for durable replication. If direct delivery is unavailable, use only the configured allowed fallback. If no permitted path exists, report Unavailable.

Fallback availability depends on the configured target and paths. Measure latency against the actual topology, encoding, and transport before assigning a bound.

Manage wire priority separately from peer connection-slot priority. A high-priority notification can precede lower-priority content on one pipe without receiving a new connection slot.

Integrate an external protocol

An Atlas peer exchanges Atlas protocols. ATAK, a serial sensor, or another external system requires an adapter that owns its protocol, parsing, translation, credentials, and delivery policy. Atlas transport does not automatically speak Cursor on Target or any other external protocol.

Custom transport and protocol-handler traits are Rust extension points. Generated bindings expose managed operations, not arbitrary Rust callbacks. The Platform Rust reference defines extension contracts.