Claude Code Plugins

Community-maintained marketplace

Feedback

vastai-api

@rawveg/skillsforge-marketplace
3
0

Vast.ai API Documentation - Affordable GPU Cloud Marketplace

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name vastai-api
description Vast.ai API Documentation - Affordable GPU Cloud Marketplace

Vastai-Api Skill

Comprehensive assistance with the Vast.ai API for managing GPU instances, machine operations, and automating AI/ML workflows. This skill provides access to official documentation for programmatically controlling the Vast.ai platform.

When to Use This Skill

This skill should be triggered when working with:

  • GPU Instance Management: Creating, destroying, starting, stopping, or managing GPU instances
  • Machine Operations: Listing machines for rent, setting pricing, managing maintenance windows
  • SSH & Authentication: Managing SSH keys, API keys, and secure connections to instances
  • Billing & Credits: Viewing invoices, earnings, deposits, or transferring credits
  • Network Volumes: Creating, listing, or managing network storage volumes
  • Serverless Endpoints: Working with Vast.ai serverless workergroups and endpoints
  • Data Transfer: Copying data between instances or cloud services
  • Account Management: Managing subaccounts, environment variables, or team roles
  • CLI Operations: Using the vastai command-line tool
  • API Integration: Building applications that integrate with Vast.ai's REST API

Quick Reference

Creating and Managing Instances

Create a New GPU Instance

# Create instance from an offer
vastai create instance <offer_id> \
  --image pytorch/pytorch:latest \
  --disk 50 \
  --ssh

List Your Active Instances

# Show all instances
vastai show instances

# Show specific instance details
vastai show instance <instance_id>

Manage Instance State

# Stop an instance (pause GPU billing, storage still charged)
vastai stop instance <instance_id>

# Start a stopped instance
vastai start instance <instance_id>

# Reboot instance without losing GPU priority
vastai reboot instance <instance_id>

# Destroy instance permanently (irreversible)
vastai destroy instance <instance_id>

SSH Key Management

Add SSH Key to Account

# Add your public SSH key
vastai create ssh-key "ssh-rsa AAAAB3NzaC1yc2EA... user@host"

# List all SSH keys
vastai show ssh-keys

# Attach SSH key to specific instance
vastai attach ssh <instance_id> <ssh_key>

Search and Filter Offers

Search for GPU Offers

# Search with filters
vastai search offers \
  --gpu_name RTX_4090 \
  --num_gpus 2 \
  --disk_space 100

# Order by price
vastai search offers --order dph_total

Environment Variables (Secrets)

Manage Environment Variables

# Create encrypted environment variable
vastai create env-var MY_API_KEY "secret_value_here"

# List all environment variables
vastai show env-vars

# Update existing variable
vastai update env-var MY_API_KEY "new_secret_value"

# Delete environment variable
vastai delete env-var MY_API_KEY

Billing and Credits

View Billing Information

# Show invoices
vastai show invoices

# Show earnings (for hosts)
vastai show earnings

# Show deposit for specific instance
vastai show deposit <instance_id>

# Transfer credits to another user
vastai transfer credit recipient@email.com 25.00

Instance Logs

Retrieve Container Logs

# Get last 100 lines of logs
vastai show logs <instance_id> --tail 100

# Filter logs with grep pattern
vastai show logs <instance_id> --filter "error"

# Get daemon system logs
vastai show logs <instance_id> --daemon-logs

Data Transfer Operations

Copy Between Instances

# Copy from one instance to another
vastai copy <src_id> <dst_id> /source/path /destination/path

# Cloud copy using rclone
vastai cloud copy <instance_id> remote:bucket/path /local/path

Machine Management (for Hosts)

List Your Machine for Rent

# List machine with pricing
vastai list machine <machine_id> \
  --price_gpu 0.50 \
  --price_disk 0.10

# Unlist machine (stop renting)
vastai unlist machine <machine_id>

# Schedule maintenance window
vastai schedule maint <machine_id> \
  --sdate "2025-11-01T10:00:00" \
  --duration 3600

Key Concepts

Instance Types

  • On-Demand Instances: Pay-as-you-go GPU instances you create and manage
  • Interruptible Instances: Lower-cost instances that can be reclaimed by hosts
  • Reserved Instances: Pre-paid instances with usage discounts (up to 40%)

Pricing Model

  • GPU Pricing: Charged per hour while instance is running
  • Storage Pricing: Charged for disk space even when instance is stopped
  • Network Transfer: Upload/download bandwidth costs
  • Discounts: Available through prepayment on reserved instances

Instance States

  • starting: Instance is initializing
  • running: Instance is active and billable
  • stopped: Container stopped (storage still billable)
  • exited: Container exited or failed
  • rebooting: In process of restarting
  • recycling: Being destroyed and recreated from fresh image

Authentication

  • API Keys: Used for programmatic access via REST API
  • SSH Keys: For secure shell access to running instances
  • Environment Variables: Encrypted secrets injected into containers

Templates

Pre-configured setups containing:

  • Docker image specifications
  • Environment variables
  • Onstart scripts
  • Resource requirements
  • Port mappings

Popular templates include PyTorch, TensorFlow, Jupyter, ComfyUI, and Stable Diffusion.

Network Volumes

Shared network storage that can be:

  • Attached to multiple instances
  • Persisted independently of instance lifecycle
  • Used for datasets and model weights
  • Scaled independently

Serverless Architecture

  • Endpoints: Top-level routing and configuration
  • Workergroups: Pools of GPU instances that autoscale
  • Test Workers: Exploration phase for performance profiling
  • Target Utilization: Controls scaling behavior

Reference Files

This skill includes comprehensive documentation in references/:

llms-full.md

Complete API reference with all endpoints organized by category:

  • Accounts: API keys, SSH keys, user management, subaccounts
  • Billing: Invoices, earnings, deposits, credit transfers
  • Instances: Create, manage, destroy, reboot, logs, SSH
  • Machines: List for rent, pricing, maintenance, default jobs
  • Network Volumes: Create, list, manage shared storage
  • Search: Find offers, benchmarks, filter GPU availability
  • Serverless: Endpoints, workergroups, autoscaling configuration

Each endpoint includes:

  • HTTP method and path
  • Detailed description
  • CLI usage examples
  • Parameter specifications
  • Source documentation links

llms-txt.md

Focused documentation covering:

  • Serverless workergroup parameters and configuration
  • Endpoint management
  • QuickStart guide with setup instructions
  • Common questions and answers
  • Schema.org structured data for better searchability

llms.md

Curated list of all API operations with brief descriptions and CLI examples, organized by category for quick lookup.

Working with This Skill

For Beginners

Start here:

  1. Review the QuickStart section in llms-txt.md
  2. Follow the 4-step setup process (signup, add credit, prepare SSH, create instance)
  3. Try the basic examples in Quick Reference above
  4. Learn about instance states and pricing model in Key Concepts

First tasks to try:

  • Create an API key for authentication
  • Add your SSH public key to your account
  • Search for available GPU offers
  • Create your first instance with a template

For Intermediate Users

Focus on:

  • Environment variable management for secrets
  • Data transfer between instances and cloud storage
  • Instance lifecycle management (stop/start/reboot vs destroy)
  • Billing optimization with reserved instances
  • Custom template creation for your workflows

Useful patterns:

  • Set up auto-billing to avoid instance interruptions
  • Use environment variables for API keys and credentials
  • Schedule regular backups with copy commands
  • Monitor costs with invoice and earnings endpoints

For Advanced Users

Advanced topics:

  • Serverless endpoint and workergroup configuration
  • Machine hosting and marketplace optimization
  • Network volume architecture for shared datasets
  • Team and subaccount management
  • API integration in custom applications
  • Automated scaling strategies

Power user tips:

  • Use filter operators in search (eq, neq, gt, lt, gte, lte, in, nin)
  • Leverage launch_args for advanced instance customization
  • Implement monitoring and alerting via logs API
  • Optimize costs with bid price adjustments
  • Build workflows with cloud copy for data pipelines

Navigation Tips

Finding API endpoints:

  • All endpoints documented in llms-full.md with full details
  • Organized by category (accounts, billing, instances, machines, etc.)
  • Each includes CLI usage examples

Quick lookups:

  • llms.md provides condensed list of all operations
  • Use browser search (Ctrl+F) to find specific commands
  • Look for "CLI Usage:" sections for command syntax

Understanding concepts:

  • Key Concepts section above for terminology
  • QuickStart in llms-txt.md for getting started
  • Workergroup Parameters section for serverless configuration

Common Workflows

Setting Up a New Development Environment

  1. Create API key with appropriate permissions
  2. Add SSH key to account for access
  3. Create environment variables for secrets
  4. Search for GPU offers matching requirements
  5. Create instance from template
  6. Connect via SSH and verify setup

Managing Long-Running Training Jobs

  1. Create instance with sufficient disk space
  2. Set up auto-billing to prevent interruptions
  3. Use reserved instance with prepayment for discounts
  4. Monitor with logs endpoint
  5. Copy model checkpoints to cloud storage
  6. Stop (not destroy) when paused to save costs

Hosting Machines for Profit

  1. Set machine pricing with list command
  2. Define minimum bid thresholds
  3. Configure default jobs for background work
  4. Schedule maintenance windows when needed
  5. Monitor earnings and clean up expired contracts
  6. Adjust pricing based on market conditions

Best Practices

Cost Management

  • Destroy vs Stop: Use stop for short pauses, destroy for long breaks
  • Disk Space: Choose carefully - cannot be changed later
  • Reserved Instances: Prepay for 40% discount on long-running work
  • Auto-billing: Set threshold above daily spend to prevent interruptions
  • Low Balance Alerts: Enable email notifications as backup

Security

  • API Keys: Use permission scoping, rotate regularly
  • SSH Keys: Use different keys for different purposes
  • Environment Variables: Store secrets as encrypted env vars
  • Subaccounts: Use for team members with restricted access

Performance

  • Template Caching: Pre-pulled images start much faster (seconds vs minutes)
  • Network Volumes: Use for large datasets shared across instances
  • Bid Pricing: Higher bids get better hardware availability
  • Test Workers: Let serverless explore before scaling

Reliability

  • Logs: Monitor regularly for errors
  • Health Checks: Implement in your applications
  • Data Backup: Copy critical data off instances regularly
  • Redundancy: For critical work, run on multiple instances

Resources

Official Links

Support

  • Minimum deposit: $5
  • Balance shown at top right of dashboard
  • Email verification required to rent or create teams
  • Auto-billing prevents interruptions when configured

Notes

  • This skill was automatically generated from official Vast.ai documentation
  • Reference files preserve structure and examples from source docs
  • CLI examples use the vastai command-line tool
  • API endpoints support both REST API and CLI access
  • All prices in USD, billed per hour for compute and storage

Updating

To refresh this skill with updated documentation:

  1. Re-run the scraper with the same configuration
  2. The skill will be rebuilt with the latest API information
  3. Check for API version changes or deprecated endpoints