What problem does it solve? Running GPU workloads requires provisioning infrastructure, handling authentication, and monitoring session lifecycles. This Skill lets you launch, monitor, and terminate GPU compute sessions on the Mimiry softlaunch platform directly from your agent, without manually wiring up API calls or SSH signature authentication. ## Core Features & Use Cases - Session Lifecycle Management: Create GPU sessions with PyTorch, TensorFlow, CUDA, or custom container images, poll until running, view logs, and terminate when done. - SSH Signature Authentication: Authenticate with the Mimiry API using your registered SSH key via the bundled auth script, with token caching and auto-refresh. - GPU Availability & Pricing: Query the public availability endpoint to find GPU models, providers, locations, and hourly rates before creating a session. - Persistent Block Volumes: Create, resize, attach, and delete block volumes that survive session termination. - Use Case: Ask your agent to launch a PyTorch training job on the cheapest available GPU — it authenticates, creates the session, polls until running, and prints SSH and management commands you can paste into your terminal. ## Quick Start Ask your agent to check your Mimiry balance and list running sessions using your SSH key at ~/.ssh/mimiry.
