Vibesdk
An open-source vibe coding platform that helps you build your own vibe-coding platform, built entirely on Cloudflare stack
Last verified:
What is Vibesdk?
VibeSDK is an open-source AI vibe coding platform built entirely on Cloudflare's developer stack that lets users build applications by describing what they want in natural language. Once deployed, users type what they want to build, and an AI agent generates, previews, and deploys the application automatically. It's designed for companies building AI-powered platforms, internal development teams, and SaaS platforms wanting to let customers extend their product's functionality without coding.
Key features include phase-wise AI code generation with intelligent error correction, live previews running in sandboxed Cloudflare Containers, interactive chat for guiding development through conversation, and generation of modern React + TypeScript + Tailwind apps. The platform supports one-click deployment to Workers for Platforms, GitHub integration for exporting code to repositories, and includes a TypeScript SDK for programmatic access.
The platform is built using Cloudflare's full ecosystem: React + Vite frontend, Workers with Durable Objects for AI agents, D1 SQLite database with Drizzle ORM, AI Gateway for multiple LLM providers, sandboxed containers for previews, R2 for storage, and Workers for Platforms with dispatch namespaces for deployment. It includes pre-built templates to speed up development and implements enterprise-grade security with encrypted secrets, sandboxed execution, input validation, rate limiting, and audit logs.
VibeSDK is perfect for startups wanting to enter the AI development space, established companies adding AI capabilities to developer tools, non-technical teams creating tools without engineering resources (marketing building landing pages, sales creating dashboards), and SaaS platforms enabling customer customizations without API learning curves.
Vibesdk pricing
Pricing model: Freemium
VibeSDK itself is open-source (MIT License) and free to use, but requires paid Cloudflare infrastructure: Cloudflare Workers Paid Plan, Workers for Platforms subscription, and Advanced Certificate Manager add-on for first-level subdomains. A Google Gemini API key from ai.google.dev is required (paid API usage). Sandbox instance types range from lite (256 MiB, free tier available) to standard-3 (12 GiB, 2 vCPU, default for production) and standard-4 (12 GiB, 4 vCPU, high-performance). AI Gateway provides caching to reduce inference costs across OpenAI, Anthropic, and Google models. Costs include Cloudflare subscription fees, API usage fees for LLM providers, and container resource usage based on selected instance type.
Vibesdk pros
- Open-source and self-hosted - complete control over infrastructure and data
- One-click deployment to Cloudflare with automated setup
- Full-stack AI webapp generator from natural language prompts
- Phase-wise code generation with intelligent error correction
- Live previews in secure sandboxed Cloudflare Containers
- Interactive chat interface for natural language development guidance
- Generates modern React + TypeScript + Tailwind applications
- One-click deploy to Workers for Platforms
- GitHub integration for direct code export to repositories
- TypeScript SDK available for programmatic programmatic access
- Built-in AI Gateway with multi-model support (OpenAI, Anthropic, Google)
- Automatic caching of popular responses to reduce inference costs
- Complete observability into token usage, costs, and performance
- Pre-built project templates to speed up development
- Encrypted secrets storage with Cloudflare encryption
- Sandboxed execution prevents untrusted code from affecting infrastructure
- Input validation and content filtering for AI-powered safety
- Audit logs track all generation activities
- Scalable to thousands or millions of deployed applications
- Customizable AI behavior and code generation patterns
Vibesdk cons
- Requires Cloudflare Workers Paid Plan (not available on free tier)
- Workers for Platforms subscription required for deployment
- Advanced Certificate Manager add-on needed for first-level subdomains
- Manual DNS setup required - wildcard CNAME record must be configured
- Google Gemini API key required as mandatory prerequisite
- OAuth setup is not automatic - must be configured manually after deployment
- DNS propagation can take up to an hour before previews resolve
- Custom domain configuration is required (not optional)
- Sandbox instance type affects performance and must be selected consciously
- Not suitable for completely non-technical users without any setup knowledge
- CI/CD requires cloning repository and manual .prod.vars configuration
- Local development requires Bun installation and multiple environment variables
- AI Gateway must be manually created if not using automatic setup
- Container resource limits may require upgrading for complex applications
- GitHub Export requires separate OAuth app configuration from login OAuth
Frequently asked questions about Vibesdk
What is VibeSDK and what does it do?
VibeSDK is an open-source AI vibe coding platform that lets users build applications by describing what they want in natural language. Users type their app idea, and an AI agent generates the code, provides live previews in sandboxed containers, and deploys the application to Cloudflare Workers. It generates React + TypeScript + Tailwind applications with phase-wise development, intelligent error correction, and interactive chat for refinements.
How do I deploy VibeSDK?
Click the 'Deploy to Cloudflare' button at build.cloudflare.dev. You need a Cloudflare Workers Paid Plan, Workers for Platforms subscription, Advanced Certificate Manager for first-level subdomains, and a Google Gemini API key. The deployment automatically creates a GitHub repository in your account, provisions the worker, and sets up required resources. After deployment, configure environment variables including GOOGLE_AI_STUDIO_API_KEY, JWT_SECRET, WEBHOOK_SECRET, SECRETS_ENCRYPTION_KEY, and CUSTOM_DOMAIN.
What are the sandbox instance types and which should I choose?
VibeSDK offers five container instance types: lite (256 MiB, 1/16 vCPU, 2 GB disk, for development/testing), standard-1 (4 GiB, 1/2 vCPU, 8 GB, for light production), standard-2 (8 GiB, 1 vCPU, 12 GB, for medium workloads), standard-3 (12 GiB, 2 vCPU, 16 GB, default for production apps), and standard-4 (12 GiB, 4 vCPU, 20 GB, for high-performance apps). Standard-3 is recommended for the best balance of performance and resources.
How does the AI code generation work?
VibeSDK generates apps in intelligent phases: Planning Phase analyzes requirements and creates file structure, Foundation Phase generates package.json and setup files, Core Phase creates main components and logic, Styling Phase adds CSS and visual design, Integration Phase connects APIs and external services, and Optimization Phase improves performance and fixes errors. The AI captures console output, build logs, and error messages to feed back for automatic fixes.
Can I use LLM providers other than Google Gemini?
Yes, VibeSDK is set up with AI Gateway that provides unified access to multiple LLM providers including OpenAI, Anthropic, Google, and others. By default it uses Google's Gemini models (gemini-2.5-pro, gemini-2.5-flash-lite, gemini-2.5-flash) for project planning, code generation, and debugging, but AI Gateway enables routing across providers, caching popular responses, and observability into costs and performance.
How do I export my generated application?
VibeSDK supports one-click project export to your Cloudflare account or GitHub repository. For GitHub export, you need to set up a separate GitHub OAuth app (different from the login OAuth app) with callback URL https://your-worker-name.workers.dev/api/github-exporter/callback. Add GITHUB_EXPORTER_CLIENT_ID and GITHUB_EXPORTER_CLIENT_SECRET to .dev.vars and .prod.vars, then redeploy.
What security measures does VibeSDK implement?
VibeSDK implements enterprise-grade security including encrypted secrets storage with Cloudflare encryption, completely isolated sandboxed containers for generated apps, input validation and sanitization for all user inputs, rate limiting to prevent abuse, AI-powered content filtering for inappropriate content detection, and complete audit logs tracking all generation activities. Each user gets their own isolated container that cannot affect other users or infrastructure.
How do I set up OAuth for user login?
OAuth is not shown on initial deployment and must be configured after deployment. For Google OAuth: create a Google Cloud Console project, enable Google+ API, create OAuth 2.0 Client ID, add authorized origins and redirect URI, then add GOOGLE_CLIENT_ID and GOOGLE_CLIENT_SECRET to both .dev.vars and .prod.vars. For GitHub OAuth: create a new OAuth App in GitHub Settings → Developer settings, set homepage and callback URLs, then add GITHUB_CLIENT_ID and GITHUB_CLIENT_SECRET to both files and redeploy.
Can I run VibeSDK locally?
Yes, clone the repository with git clone https://github.com/cloudflare/vibesdk.git, cd into it, run npm install or bun install, then run npm run setup or bun run setup. The setup script guides you through installing Bun, configuring Cloudflare credentials, setting up AI providers and OAuth, database setup, and template deployment. Start the development server with bun run dev and visit http://localhost:5173.
What templates are available and how can I add more?
VibeSDK includes pre-built project templates stored in an R2 bucket that the LLM uses as starting points for common applications to speed up development. The official vibesdk-templates repository contains the templates catalog powering VibeSDK. You can expand the template set and add more examples by adding templates to your R2 bucket. Templates help the AI quickly customize existing patterns instead of generating every file from scratch.