Vast.ai Skill Pack

Claude Code skill pack for Vast.ai GPU cloud marketplace integration (24 skills)

Installation

/plugin install vastai-pack@claude-code-plugins-plus

About Vast.ai

Vast.ai is a GPU cloud marketplace where individual hosts and data centers list GPU machines at prices 50-90% below hyperscaler providers. Common use cases: ML training, inference, rendering, and any GPU-accelerated workload where cost efficiency matters.

Key facts:

• REST API at cloud.vast.ai/api/v0 + CLI via pip install vastai
• Pricing: RTX 4090 ~$0.15-0.30/hr, A100 ~$1.00-2.00/hr, H100 ~$2.50-4.00/hr
• Instance types: on-demand and interruptible (spot)
• Access: SSH + Docker containers on rented GPU hardware

Skills Included

Standard Skills (S01-S12)

| Skill | Description | |-------|-------------| | vastai-install-auth | Install CLI, configure API key, build Python client | | vastai-hello-world | Rent first GPU instance, run PyTorch workload, destroy | | vastai-local-dev-loop | Mock API, test Docker images locally, connection testing | | vastai-sdk-patterns | Typed queries, context-managed lifecycle, offer scoring | | vastai-core-workflow-a | Search, provision, execute job, collect artifacts, destroy | | vastai-core-workflow-b | Multi-instance orchestration, spot recovery, cost analysis | | vastai-common-errors | API errors, instance failures, SSH issues, CUDA problems | | vastai-debug-bundle | Account diagnostics, instance logs, GPU health, nvidia-smi | | vastai-rate-limits | Rate-limited client, adaptive polling, request optimization | | vastai-security-basics | API key management, SSH hardening, data cleanup | | vastai-prod-checklist | Production readiness audit with verification script | | vastai-upgrade-migration | CLI upgrades, CUDA migration, Docker image updates |

Pro Skills (P13-P18)

| Skill | Description | |-------|-------------| | vastai-ci-integration | GitHub Actions GPU testing with budget controls | | vastai-deploy-integration | Automated deployment scripts, Docker optimization | | vastai-webhooks-events | Instance lifecycle polling, auto-recovery handlers | | vastai-performance-tuning | dlperf scoring, DataLoader tuning, batch sizing | | vastai-cost-tuning | GPU cost-efficiency, spot vs on-demand, auto-destroy | | vastai-reference-architecture | Three-tier architecture, checkpoint manager |

Flagship Skills (F19-F24)

| Skill | Description | |-------|-------------| | vastai-multi-env-setup | Dev/staging/prod API keys, GPU whitelists, budgets | | vastai-observability | Metrics collection, GPU alerts, Prometheus exporter | | vastai-incident-runbook | Triage scripts, spot recovery, billing emergency stop | | vastai-data-handling | Encrypted transfer, cloud checkpoints, secure cleanup | | vastai-enterprise-rbac | Team budgets, policy enforcement, audit logging | | vastai-migration-deep-dive | AWS/GCP to Vast.ai, cost comparison, Docker adaptation |

Quick Start

# Install CLI
pip install vastai

# Configure API key
vastai set api-key YOUR_KEY_FROM_CLOUD_VAST_AI

# Search for cheap GPUs
vastai search offers 'num_gpus=1 gpu_ram>=24 reliability>0.95' --order dph_total --limit 5

# Rent an instance
vastai create instance OFFER_ID --image pytorch/pytorch:2.2.0-cuda12.1-cudnn8-runtime --disk 20

# Connect and verify
ssh -p PORT root@HOST "nvidia-smi"

# IMPORTANT: Destroy when done (stops billing)
vastai destroy instance INSTANCE_ID

Resources

• Vast.ai Documentation
• REST API Reference
• CLI GitHub
• Search & Filter

License

MIT