VibeForge: Self hosted AWS Vibe coding platform

An AI agent that turns a prompt into a fully deployed AWS app—real infra, no lock-in, powered by Bedrock and AgentCore for true “vibe coding” inside your own AWS cloud.

Comment

Inspiration

I love the idea of “vibe coding” — describing what you want and watching it come to life. But most existing platforms keep you locked into their ecosystem or charge for hosting. I wanted that same creative, instant experience while keeping everything inside my own AWS account. That’s what inspired VibeForge — a self-hosted, AWS-native way to turn prompts into full-stack apps you actually own.

What it does

VibeForge is a local AI agent that converts natural language prompts into real, deployable AWS applications. It uses Amazon Bedrock for reasoning, generates code and infrastructure with AWS CDK, and deploys automatically to dev and prod using Lambda, API Gateway, DynamoDB, and CloudFront — all inside the user’s AWS account. It’s like having your own private “AI coding assistant” that builds directly in your cloud.

How we built it

I built VibeForge using:

  • Amazon Bedrock + AgentCore for the reasoning and planning flow
  • AWS CDK for infrastructure-as-code and deployments
  • STS AssumeRole + CloudFormation Quick-Create for safe, one-click account access
  • Next.js and Node.js for the self-hosted “vibe coding” UI and control service

Challenges we ran into

As a solo builder, the hardest part was connecting everything securely while keeping it simple for users. Getting Bedrock access configured in multiple regions, handling IAM boundaries for CDK, and dealing with CloudFormation permission errors took a lot of trial and error. Balancing automation with transparency — so users always know what’s deploying into their account — was also a key design challenge.

Accomplishments that we're proud of

  • Built a working prompt-to-deploy pipeline that runs completely in the user’s AWS account
  • Achieved zero vendor lock-in — all generated code and infra belong to the user
  • Simplified onboarding to a single CloudFormation click, no CLI or AWS keys required

What we learned

Working alone on VibeForge taught me a lot about combining AI reasoning with real infrastructure. I learned how to use Amazon Bedrock effectively for planning and code generation, how to structure CDK stacks for multi-environment deployments, and how small IAM misconfigurations can break automation in subtle ways. I also learned that great developer experience doesn’t have to mean giving up control — you can keep things open, secure, and still feel magical.

What's next for VibeForge: Self-hosted AWS Vibe coding platform

  • Add multi-agent collaboration using Bedrock AgentCore and Amazon Q
  • Expand app templates for containers, APIs, and data workflows
  • Support AWS SageMaker integration for agent-driven ML backends

Built With

Share this project:

Updates