CmdStan

CmdStan is the command-line interface for the Stan probabilistic programming language. It provides the core workflow for compiling Stan models and running inference from the shell, making it a foundational tool in the Stan ecosystem. CmdStan is designed for scripting, automation, and use on various computing environments, including high-performance systems.

A typical workflow starts with writing a Stan model in a .stan file. The model is compiled into an executable, after which data are supplied (commonly in JSON or R dump format) and a run is launched via the command line. CmdStan supports different inference methods, including sampling with Markov chain Monte Carlo, as well as optimization and variational inference. When performing sampling, users can configure the number of chains, warmup iterations, total iterations, thinning, and seeds. CmdStan can launch multiple chains in parallel, leveraging available CPU cores. Output includes posterior samples in a draws file (CSV) and accompanying diagnostics data, along with log messages and related files.

CmdStan is cross-platform, compatible with Linux, macOS, and Windows environments that provide a compatible toolchain. It relies on a C++ compiler and build tools, and uses a straightforward, script-friendly workflow that emphasizes reproducibility and automation through shell scripts or integration into larger pipelines.

In the broader Stan ecosystem, CmdStan complements higher-level interfaces such as PyStan and CmdStanPy for Python users and RStan for R users. CmdStan remains a lightweight, language-agnostic option for running Stan models, offering detailed control over model compilation and inference without additional language bindings.