rustsim - high-performance agent-based modelling engine.

This is the top-level orchestration crate that re-exports all sub-crates and adds:

• [compute] - CUDA/CPU compute routing and SoA batch stepping.
• [control_plane] - deterministic replay, checkpoint/restore, engine metrics, typed failures, and partition metadata.
• [device_store] - persistent device-side SoA storage (FlameGPU2-style).
• [ensemble] - parameter sweep and batch simulation runner.
• [layers] - layer-based execution model for multi-function steps.
• [interaction] - columnar spatial interaction/message indexes.
• [telemetry] - end-to-end pipeline from data collection to ClickHouse.

Quick start

See the full working example under examples/basic_particle.rs:

use rand::rngs::StdRng;
use rand::SeedableRng;
use rustsim::prelude::*;
use rustsim_spaces::nothing::NothingSpace;

#[derive(Debug, Clone)]
struct Particle { id: AgentId, x: f64 }
impl Agent for Particle { fn id(&self) -> AgentId { self.id } }

let mut store: HashMapStore<Particle> = HashMapStore::new();
store.insert(Particle { id: 1, x: 0.0 });

let rng = StdRng::seed_from_u64(42);
let mut model: StandardModel<NothingSpace, Particle, _, (), _, Fastest> =
    StandardModel::new(
        store,
        NothingSpace,
        Fastest::new(),
        (),
        rng,
        None::<Box<dyn FnMut(&mut Particle, &mut StepContext<NothingSpace, Particle, (), StdRng, Fastest>)>>,
        None,
        true,
    );
model.step_n(100);

Crate architecture