Crate helixrouter 

HelixRouter

Adaptive async compute routing engine for Rust.

HelixRouter decides how work runs — inline, spawned, pooled, batched, or dropped — on a per-job basis in sub-microsecond time, using live system pressure, EMA latency history, and an online-learned quality model (neural_router).

Modules

Module	Purpose
router	Core strategy selection and execution dispatch
neural_router	Online-learning per-job-kind routing quality model
autoscaler	Predictive demand forecasting and capacity recommendations
config	Validated RouterConfig, hot-reload, watch channel
metrics	EMA latency, percentiles, pressure scoring, Prometheus export
strategies	Deterministic CPU-bound compute kernels
simulator	Seeded synthetic workload generation
web	Axum HTTP server, SSE feed, embedded dark dashboard
types	Shared data types: Job, Strategy, Output

Quick start

use helixrouter::{config::RouterConfig, router::Router, types::{Job, JobKind}};

#[tokio::main]
async fn main() {
    let router = Router::new(RouterConfig::default());
    let job = Job {
        id: 1,
        kind: JobKind::HashMix,
        inputs: vec![42],
        compute_cost: 1_000,
        scaling_potential: 0.5,
        latency_budget_ms: 50,
    };
    let output = router.submit(job).await;
    println!("{output:?}");
}

Modules

autoscaler

Stage: Autoscaler

config

Module: config

metrics

Module: metrics

neural_router

Module: NeuralRouter

router

Module: router

simulator

Simulator module is only available when the simulation feature is enabled.

strategies

Module: strategies

types

Core data types shared across all modules.

web

Module: web