Polyaxon

An AI control plane for Kubernetes that schedules, tracks and governs GPU training, inference, agent sessions and related workloads on customer-owned clusters and storage. It addresses running and operating machine learning and AI workloads with experiment tracking, orchestration, registries and observability in one workflow. It is for developers, data teams and operations teams in businesses that run AI engineering on Kubernetes. It is delivered as an Apache 2.0 open-source core with self-hosted and cloud-connected deployment options.

Key features

• GPU training and fine-tuning
• Distributed multi-node training
• Hyperparameter sweeps
• DAGs and batch jobs
• Agent workload scheduling
• Model serving and inference
• Interactive notebooks and sandboxes
• Experiment tracking and comparison
• Dashboards and observability
• Model registry and lineage
• Evaluation and AI gateway
• Scheduling and orchestration queues
• Governance and access controls
• Audit logs and policies
• Automation workflows
• CLI API and SDKs