Forged for Local Compute & Hardware Truth
Crisol turns your heterogeneous workstations, Apple Silicon MacBooks, NVIDIA RTX rigs, and cloud pods into a unified, deterministic compute forge. Zero bloat. Pure reproducibility.
Built Against Speculation. Grounded in Silicon.
Universal Job Dispatch
One standardized Python runner (`crisol.py run <job>`) routes workloads seamlessly to the optimal node—whether it's an M-series Mac via SSH, an RTX rig via CUDA, or a remote cloud cluster.
Deterministic Audit Trail
Every benchmark enforces identical seeds, matched dimensions, warmup iterations, and PyTorch backend constraints. Full JSONL telemetry records exact TFLOPS, memory, and exit codes.
Zero-Bloat Cluster Harness
No heavy Kubernetes, Slurm, or Ray configurations. Crisol connects your existing desktop, laptop, and server hardware over native, lightweight transport with zero runtime friction.
The Silicon Forge for Every Developer
Stop letting compute sit idle in different rooms. Crisol unites your local silicon into a reproducible, unified laboratory. Open source core, community-driven, built for the next generation of AI systems.