KairoECS Documentation
Build reproducible discrete-event and agent-based models in Rust.
KairoECS brings event scheduling, entity/component state, resource queues and agent behaviours into one modular engine. Optional Rust crates provide columnar data exchange, parallel execution, distributed synchronization and GPU kernels. Applications own their domain models and depend on Kairos through versioned interfaces.
Start here
Section titled “Start here”- Rust quickstart — execute a small scheduler example.
- Architecture — understand state, time and adapter boundaries.
- Installation — build the pinned Rust source and select a native adapter.
- Docs platform and Learning Coverage Matrix — documentation workflow and learning material boundaries.
- Capabilities and evidence — see what has been implemented, executed and independently checked.
Alpha, with one simulation core
Section titled “Alpha, with one simulation core”v0.4.0-alpha.1 was published on 18 October 2026. Default-branch work is experimental, with explicit beta and stable-release gates.
Rust is the only supported simulation core. Python, R, Julia, TypeScript, C# and Go are adapters to native Rust or Rust-compiled WebAssembly. A missing native operation returns an error; a host-language engine is never substituted. Former pure-language preview cores are unsupported immediately. See the compatibility notice before using an earlier quickstart or package.
Capabilities at a glance
Section titled “Capabilities at a glance”| Area | What to expect |
|---|---|
| Core DES and ABM | Ordered scheduling, cancellation, generational state, resource primitives, spatial behaviours and versioned RNG streams. |
| Data exchange | Optional Arrow/Polars and native ownership interfaces; qualify the installed consumer and enabled features. |
| Parallel and distributed | Implemented conservative and optimistic PDES, native MPI and typed gRPC, with bounded execution and recovery evidence. |
| GPU | Specific device kernels and persistent-buffer paths; general model offload and end-to-end speedup are not established. |
| Scientific extensions | Uneven maturity: named numerical methods, inference, RL and third-party FMI workflows have explicit completion gates. |
The capability register records the scope of each claim. Track completion is not interchangeable with scientific validation or release acceptance.
Evaluate before scaling
Section titled “Evaluate before scaling”The measured conservative-PDES benchmark reports slower execution than the sequential baseline on its small workloads. The corrected scaling report separates generated fixtures from measurements. No general performance-leadership or HPC scaling claim is currently supported by those fixtures.
For research and operational models, use the validation dataset and oracle plan, record uncertainty and check predictions against independent data.
Continue learning
Section titled “Continue learning”- Language adapters, TypeScript/Wasm and the binding inventory.
- PDES and distributed evidence, GPU evidence and Conductor status.
- Tutorials and examples, community and trustworthy simulation.
- Citation, release checklist and development programme.
Documentation workflow
Section titled “Documentation workflow”The active site is Astro/Starlight. Source documentation remains in docs/,
bindings/, examples/ and conductor/. The
docs platform explains versioned pages, compatibility routes,
search and llms.txt.
Run just docs-build to build the site and just docs-dev to start the local
documentation server. The contributor workflow
explains prerequisites and validation commands.
npm --prefix website cinpm --prefix website run check:all