Trainy

Trainy is a GPU workflow suite for AI teams, covering training job scheduling, inference serving and experiment tracking. Teams can adopt one product or all three, and each is built around keeping long running GPU work reliable. Konduktor handles training and inference. A job is described in a YAML file with node count, accelerator type and priority, then launched with a single command, and Konduktor sets up the distributed run. It detects bad GPUs and nodes, cordons them and reschedules the job onto healthy hardware, and it provides aggregated logs and diagnostics, priority scheduling with preemption, RDMA native multi node training, multi cloud job queuing and observability through Prometheus and Grafana. It deploys inside the customer's own cloud account or on bare metal. Inference endpoints add autoscaling, load balancing and health checks. Pluto is the experiment tracker, with Weights and Biases and Neptune compatible APIs, dual logging for migration and an MCP integration. The audience is machine learning and infrastructure teams running training at scale. Konduktor is sold on a custom quote sized to the cluster, with a platform fee plus per node pricing. Pluto has a free hosted tier with 2 seats and 2 GB of storage, a Pro tier at $250 per seat per month that includes up to 10 seats, and an enterprise tier above that. Pluto is open source and can be self hosted for free, and it is free for academics with a .edu email address.