Graph representation:
Phasic represents Markov jump processes as graphs rather than matrices. User models are specified with a simple callback function.

Moments and Distributions:
Compute PDFs and any number of exact moments for both continuous to discrete phase-type distributions.

Bayesian Inference:
Parameter inference using Stein Variational Gradient Descent produce confidence intervals of estimates.

Graph algorithms:
Efficient graph-based algorithms directly implemented in C with APIs for both Python, R, and C++. Tracing and caching of symbolic graph operations in Gaussian elimination allow evaluation of large models across parameter values.

Multiprocessing:
Automatically parallelizes inference computation across CPUs on a single device or multiple devices on a SLURM cluster.

Distributed symbolic elimination:
Symbolic elimination traces are cached for reuse and community sharing by IPFS and a GitHub model hub.