MLOS Foundation is building the operating layer between your ML models and your hardware.
We develop open-source ML infrastructure — starting with a purpose-built Machine Learning Operating System featuring kernel-level optimizations for ML workloads, and Axon, a universal model installer that works across HuggingFace, PyTorch Hub, TensorFlow Hub, and ModelScope.
What we're solving: Today, deploying ML models means wrestling with fragmented toolchains, incompatible frameworks, and manual optimization. MLOS provides a unified runtime and model lifecycle management layer — one CLI, any model source, automatic ONNX conversion, and optimized inference.
What we've built:
→ Axon: Open-source model installer covering 80%+ of the ML ecosystem
→ MLOS Core: ML runtime with Standard Model Interface (SMI) for unified inference
→ 18+ models tested at 100% pass rate with zero inference failures
→ GPT-2 inference at 62ms, full LLM inference at 3.4–6.4s on CPU
Two USPTO provisional patents filed for kernel-level ML optimizations and universal model lifecycle management.
We believe ML infrastructure should be as standardized as operating systems made compute. No vendor lock-in. No framework wars. Just models that work.
Open source: github.com/mlOS-foundation
Validation reports: mlos-foundation.github.io/system-test
Contact: info@mlosfoundation.org

MLOSFoundation

Org chart

Teams

Offices