vastai-api

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Vastai-Api Skill

Vastai-Api 技能文档

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.

为管理GPU实例、机器操作以及自动化AI/ML工作流提供全面的Vast.ai API使用指导。本技能文档包含用于程序化控制Vast.ai平台的官方文档内容。

When to Use This Skill

何时使用本技能文档

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

在处理以下场景时可使用本技能文档：

GPU实例管理：创建、销毁、启动、停止或管理GPU实例
机器操作：列出可出租机器、设置定价、管理维护窗口
SSH与身份验证：管理SSH密钥、API密钥以及与实例的安全连接
账单与信用额度：查看发票、收益、存款或转账信用额度
网络卷：创建、列出或管理网络存储卷
无服务器端点：使用Vast.ai无服务器工作组和端点
数据传输：在实例之间或与云服务之间复制数据
账户管理：管理子账户、环境变量或团队角色
CLI操作：使用
```
vastai
```
命令行工具
API集成：构建与Vast.ai REST API集成的应用

Quick Reference

快速参考

Creating and Managing Instances

创建与管理实例

Create a New GPU Instance

创建新的GPU实例

bash

undefined

bash

undefined

Create instance from an offer

vastai create instance <offer_id>
--image pytorch/pytorch:latest
--disk 50
--ssh

undefined

vastai create instance <offer_id>
--image pytorch/pytorch:latest
--disk 50
--ssh

undefined

List Your Active Instances

列出您的活跃实例

bash

undefined

bash

undefined

Show all instances

vastai show instances

Show specific instance details

vastai show instance <instance_id>

undefined

vastai show instance <instance_id>

undefined

Manage Instance State

管理实例状态

bash

undefined

bash

undefined

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>

undefined

vastai destroy instance <instance_id>

undefined

SSH Key Management

SSH密钥管理

Add SSH Key to Account

为账户添加SSH密钥

bash

undefined

bash

undefined

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>

undefined

vastai attach ssh <instance_id> <ssh_key>

undefined

Search and Filter Offers

搜索与筛选报价

Search for GPU Offers

搜索GPU报价

bash

undefined

bash

undefined

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

undefined

vastai search offers --order dph_total

undefined

Environment Variables (Secrets)

环境变量（机密信息）

Manage Environment Variables

管理环境变量

bash

undefined

bash

undefined

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

undefined

vastai delete env-var MY_API_KEY

undefined

Billing and Credits

账单与信用额度

View Billing Information

查看账单信息

bash

undefined

bash

undefined

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

undefined

vastai transfer credit recipient@email.com 25.00

undefined

Instance Logs

实例日志

Retrieve Container Logs

获取容器日志

bash

undefined

bash

undefined

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

undefined

vastai show logs <instance_id> --daemon-logs

undefined

Data Transfer Operations

数据传输操作

Copy Between Instances

在实例之间复制数据

bash

undefined

bash

undefined

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

undefined

vastai cloud copy <instance_id> remote:bucket/path /local/path

undefined

Machine Management (for Hosts)

机器管理（针对主机方）

List Your Machine for Rent

列出可出租的机器

bash

undefined

bash

undefined

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

undefined

vastai schedule maint <machine_id>
--sdate "2025-11-01T10:00:00"
--duration 3600

undefined

Key Concepts

核心概念

Instance Types

实例类型

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%)

按需实例：您创建和管理的按需付费GPU实例
可中断实例：成本更低，但可能被主机方收回的实例
预留实例：预付费实例，可享受使用折扣（最高40%）

Pricing Model

定价模型

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

GPU定价：实例运行时按小时计费
存储定价：即使实例停止，仍会对磁盘空间计费
网络传输：上传/下载带宽费用
折扣：通过预留实例预付费可享受折扣

Instance States

实例状态

```
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

```
starting
```
：实例正在初始化
```
running
```
：实例处于活跃状态并计费
```
stopped
```
：容器已停止（仍对存储计费）
```
exited
```
：容器已退出或运行失败
```
rebooting
```
：正在重启过程中
```
recycling
```
：正在销毁并从全新镜像重新创建

Authentication

身份验证

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

API密钥：用于通过REST API进行程序化访问
SSH密钥：用于安全连接到运行中的实例
环境变量：注入到容器中的加密机密信息

Templates

模板

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.

预配置的环境设置，包含：

Docker镜像规格
环境变量
启动脚本
资源需求
端口映射

热门模板包括PyTorch、TensorFlow、Jupyter、ComfyUI和Stable Diffusion。

Network Volumes

网络卷

Shared network storage that can be:

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

可共享的网络存储，具备以下特性：

可附加到多个实例
独立于实例生命周期持久化
可用于存储数据集和模型权重
可独立扩展

Serverless Architecture

无服务器架构

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

端点：顶层路由与配置
工作组：可自动扩缩容的GPU实例池
测试工作节点：用于性能分析的探索阶段
目标利用率：控制扩缩容行为

Reference Files

参考文档

This skill includes comprehensive documentation in

references/

本技能文档包含

references/

目录下的完整文档：

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

完整的API参考文档，所有端点按类别组织：

账户：API密钥、SSH密钥、用户管理、子账户
账单：发票、收益、存款、信用额度转账
实例：创建、管理、销毁、重启、日志、SSH
机器：列出可出租机器、定价、维护、默认任务
网络卷：创建、列出、管理共享存储
搜索：查找报价、基准测试、筛选GPU可用性
无服务器：端点、工作组、自动扩缩容配置

每个端点包含：

HTTP方法与路径
详细描述
CLI使用示例
参数说明
源文档链接

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

聚焦于以下内容的文档：

无服务器工作组参数与配置
端点管理
包含设置说明的快速入门指南
常见问题与解答
用于提升搜索可见性的Schema.org结构化数据

llms.md

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

按类别组织的所有API操作精选列表，包含简要描述和CLI示例，便于快速查阅。

Working with This Skill

使用本技能文档的指南

For Beginners

面向初学者

Start here:

Review the QuickStart section in
```
llms-txt.md
```
Follow the 4-step setup process (signup, add credit, prepare SSH, create instance)
Try the basic examples in Quick Reference above
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

入门步骤：

查看
```
llms-txt.md
```
中的快速入门部分
遵循4步设置流程（注册、添加信用额度、准备SSH、创建实例）
尝试上述快速参考中的基础示例
学习核心概念中的实例状态与定价模型

推荐首次尝试的任务：

创建用于身份验证的API密钥
将您的SSH公钥添加到账户
搜索可用的GPU报价
使用模板创建您的首个实例

For Intermediate Users

面向中级用户

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

重点关注：

机密信息的环境变量管理
实例与云存储之间的数据传输
实例生命周期管理（停止/启动/重启 vs 销毁）
使用预留实例优化账单成本
为您的工作流创建自定义模板

实用模式：

设置自动账单以避免实例中断
使用环境变量存储API密钥和凭证
使用复制命令定期备份数据
通过发票和收益端点监控成本

For Advanced Users

面向高级用户

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

高级主题：

无服务器端点与工作组配置
机器托管与市场优化
用于共享数据集的网络卷架构
团队与子账户管理
在自定义应用中集成API
自动化扩缩容策略

高级用户技巧：

在搜索中使用过滤运算符（eq、neq、gt、lt、gte、lte、in、nin）
利用launch_args进行高级实例自定义
通过日志API实现监控与告警
通过调整竞价价格优化成本
使用云复制命令构建数据管道工作流

Navigation Tips

导航技巧

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

查找API端点：

所有端点的完整详情记录在
```
llms-full.md
```
中
按类别组织（账户、账单、实例、机器等）
每个端点都包含CLI使用示例

快速查阅：

```
llms.md
```
提供了所有操作的精简列表
使用浏览器搜索（Ctrl+F）查找特定命令
查找“CLI Usage:”部分获取命令语法

理解概念：

核心概念部分包含术语解释
```
llms-txt.md
```
中的快速入门指南帮助您上手
工作组参数部分介绍无服务器配置

Common Workflows

常见工作流

Setting Up a New Development Environment

设置新的开发环境

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

创建具备适当权限的API密钥
为账户添加SSH密钥以实现访问
创建用于存储机密信息的环境变量
搜索符合需求的GPU报价
使用模板创建实例
通过SSH连接并验证环境设置

Managing Long-Running Training Jobs

管理长期运行的训练任务

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

创建具备足够磁盘空间的实例
设置自动账单以防止中断
使用预付费预留实例享受折扣
通过日志端点监控任务
将模型检查点复制到云存储
暂停时停止（而非销毁）实例以节省成本

Hosting Machines for Profit

托管机器获取收益

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

使用list命令设置机器定价
定义最低竞价阈值
配置用于后台任务的默认作业
在需要时安排维护窗口
监控收益并清理过期合约
根据市场情况调整定价

Best Practices

最佳实践

Cost Management

成本管理

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

销毁 vs 停止：短期暂停使用停止，长期闲置使用销毁
磁盘空间：谨慎选择，后续无法更改
预留实例：预付费长期运行工作负载可享受40%折扣
自动账单：设置高于每日支出的阈值以避免中断
低余额告警：启用邮件通知作为备份

Security

安全

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

API密钥：使用权限范围限制，定期轮换
SSH密钥：为不同用途使用不同密钥
环境变量：将机密信息存储为加密环境变量
子账户：为团队成员创建受限访问的子账户

Performance

性能

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

模板缓存：预拉取的镜像启动速度更快（秒级 vs 分钟级）
网络卷：用于在多个实例之间共享大型数据集
竞价价格：更高的竞价可获得更好的硬件可用性
测试工作节点：让无服务器架构先探索再扩缩容

Reliability

可靠性

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

日志：定期监控以排查错误
健康检查：在您的应用中实现健康检查
数据备份：定期将关键数据复制到实例外
冗余：对于关键工作负载，在多个实例上运行

Resources

资源

Official Links

官方链接

Console: https://cloud.vast.ai/
API Docs: https://docs.vast.ai/
Postman Collection: https://www.postman.com/vast33/vast-ai-public-api-docs
Templates: https://cloud.vast.ai/templates/
Search: https://cloud.vast.ai/create/

控制台：https://cloud.vast.ai/
API文档：https://docs.vast.ai/
Postman集合：https://www.postman.com/vast33/vast-ai-public-api-docs
模板：https://cloud.vast.ai/templates/
搜索：https://cloud.vast.ai/create/

Support

支持

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

最低存款：5美元
余额显示在控制台右上角
出租机器或创建团队需要验证邮箱
配置自动账单可防止实例中断

Notes

说明

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

本技能文档由Vast.ai官方文档自动生成
参考文档保留了源文档的结构和示例
CLI示例使用
```
vastai
```
命令行工具
API端点同时支持REST API和CLI访问
所有价格以美元计价，计算资源和存储按小时计费

Updating

更新

To refresh this skill with updated documentation:

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

如需使用最新文档更新本技能：

使用相同配置重新运行抓取工具
技能文档将使用最新API信息重建
检查API版本变更或已弃用的端点