Module swarm_remote 

The agents of other AgentMux instances, for the Swarm’s sections below this instance’s own tree (docs/specs/SPEC_SWARM_OTHER_HOSTS_AND_CHANNELS_2026_10_02.md).

Phase 2 adds LAN hosts, from the LAN discovery peer list (mDNS and the UDP route): one host per hostname, one channel per advertising instance, names from each peer’s agents list.

Phase 1: other channels on this machine, read from the host-global shared registry (reactive::registry::list_all_shared), the same source the /agentmux/discovery endpoint’s host.cross_channel uses. Names only: the registry carries no status, provider or summary. LAN hosts and cloud installs are later phases and arrive as more RemoteHosts with their own tier.

Phase 4 adds each agent’s state (working, waiting, idle, …) where its instance reports one: a LAN peer’s from the agent-names poll the discovery loop already makes, another channel’s from the same route asked over loopback (channel_statuses). With it, by the owner’s decision of 2026-10-13, each agent’s summary line where its instance shares one (backend::agent_line).

Nothing here carries a credential to the frontend: the registry’s auth_key and local_url are dropped.

Structs

ChannelReport

What one channel of this machine said about its agents, by lower-cased name.

ChannelStatusTarget

One instance to ask for its agents’ states: (channel, local_url, auth_key).

RemoteAgent

RemoteChannel

RemoteHost

SwarmOtherInstancesResult

Response of swarm.other-instances.

Constants

CHANNEL_STATUS_TIMEOUT 🔒

How long another channel on this machine gets to say what its agents are doing. It is loopback, so one that takes longer is hung, and the Swarm’s poll shouldn’t wait on it: its rows show names only until it answers.

HIDE_AFTER_MS

Past this an entry is not shown at all. The registry’s own startup sweep only drops entries older than 4 h, and a crashed channel should not look present for that long.

STALE_AFTER_MS

A registry entry not rewritten for this long is shown as stale. The registry heartbeat rewrites every live agent’s entry every 20 s, so this is three missed beats.

Functions

channel_status_targets

The instances whose agents other_channels lists and that are worth asking: one per local_url, loopback only (as the cross-channel fleet forwarder requires), never this instance, and not a stale one, whose rows are dimmed and show no state.

channel_statuses

Asks each target’s agent-names route, concurrently, for its agents’ states and lines. A target that fails, times out or answers badly contributes nothing.

fnv1a32 🔒

is_loopback_url 🔒

label_repeated_channels 🔒

A host’s instances as channels, with a channel name that appears more than once suffixed by its port (“stable :29702”), so the sections stay apart.

lan_hosts

LAN hosts from the discovery peer list. A peer at one of this machine’s own addresses is another channel on this host, which the registry already lists (with more detail), so it is left out here. A peer that did not advertise its channel (an older build) is labelled by its port, so two of them on one host stay apart.

other_channels

Group the shared registry’s entries into this machine’s other channels.

snapshot

The full answer: this machine’s other channels as one host (absent when there are none), then the LAN hosts.

Type Aliases

ChannelStatuses

Each channel’s report, by channel.