land-protocol

LAND Protocol (Local AI Network Discovery)

“The HDMI of AI”

A lightweight, local-first protocol for automatic discovery and communication between AI nodes on a local network.

Why LAND?

In the current AI landscape, hardware is often siloed or cloud-dependent. LAND is an open standard that allows any device (from a tiny ESP32 to a multi-GPU server) to join a local collective intelligence with no API keys and no internet requirement.

• Discovery: mDNS on UDP port 5353.
• Transport: TCP for the cognitive API (default port 8419).

LAND separates:

• LAND (the standard): an open protocol for local AI node discovery and communication.
• LaRuche (the product): a concrete implementation of that protocol.

Core features

• Zero-config discovery: mDNS/DNS-SD peer discovery on LAN.
• Cognitive manifest: each node broadcasts identity, capabilities, and load.
• Proof of proximity: local trust/auth primitives.
• Swarm intelligence: peer state and resilience helpers.
• Priority QoS: request prioritization primitives.

How it works

Nodes advertise using _ai-inference._tcp.local. and expose key metadata through TXT records. Clients and peers listen to these announcements and build a live map of available intelligence.

Current implementation notes

• Listener stale timeout is 45s.
• ServiceRemoved events are treated as transient hints; eviction is timeout-based.
• PartialManifest::api_url() and dashboard_url() are IPv6-safe.
• URL helpers:
  • land_protocol::format_host_for_url(host)
  • land_protocol::endpoint_url(host, port, tls)
• qos::RequestQueue::dequeue() marks requests as active; call complete(qos) when processing ends.
• swarm::heartbeat() promotes Syncing, Suspect, and Down peers back to Active.

Usage

Add as dependency

[dependencies]
land-protocol = { git = "https://github.com/infinition/land-protocol" }

1. Discovering nodes (client side)

use land_protocol::discovery::LandListener;

#[tokio::main]
async fn main() {
    let mut listener = LandListener::new().unwrap();
    let nodes = listener.start().unwrap();

    tokio::time::sleep(tokio::time::Duration::from_secs(3)).await;

    let snapshot = nodes.read().await;
    for (_id, node) in snapshot.iter() {
        println!(
            "{} @ {}:{}",
            node.manifest.node_name.as_deref().unwrap_or("unknown"),
            node.manifest.host,
            node.manifest.port.unwrap_or(8419)
        );
    }
}

2. Broadcasting a node (server side)

use land_protocol::discovery::LandBroadcaster;
use land_protocol::manifest::{CognitiveManifest, HardwareTier};

#[tokio::main]
async fn main() {
    let mut manifest = CognitiveManifest::new("node-a".into(), HardwareTier::Core);
    let mut broadcaster = LandBroadcaster::new().unwrap();
    broadcaster.register(&manifest).unwrap();

    let mut interval = tokio::time::interval(tokio::time::Duration::from_secs(2));
    loop {
        interval.tick().await;
        manifest.performance.queue_depth = manifest.performance.queue_depth.saturating_add(1);
        broadcaster.update(&manifest).unwrap();
    }
}

Notes for integrators

• Prefer PartialManifest::api_url() / dashboard_url() over manual string formatting.
• Use URL helpers for direct endpoint building, especially with IPv6 hosts.
• If you use RequestQueue, pair dequeue() with complete(qos).

This site is open source. Improve this page.