J.Goutin

Posted on Dec 19 • Originally published at dev.to

Use AWS Bedrock & AI Services (Claude, Nova, Polly, Transcribe) with Your Existing OpenAI Code

#ai #devops #aws #openai

TL;DR

Use your existing OpenAI-compatible tools (OpenWebUI, n8n, Cline, Cursor) with AWS Bedrock's 80+ models and AWS AI services (Polly, Transcribe). Just change the endpoint - no code changes needed. Deploy to your AWS account in 15 minutes with Terraform. Data stays in your AWS environment for HIPAA/GDPR compliance.

The Solution in 30 Seconds

from openai import OpenAI

# Just change this line ↓
client = OpenAI(base_url="https://your-aws-api/v1", api_key="your-key")

# Everything else stays the same ↓
response = client.chat.completions.create(
    model="anthropic.claude-sonnet-4-5-20250929-v1:0",  # Claude 4.5 on Bedrock
    messages=[{"role": "user", "content": "Hello from AWS!"}]
)

What this enables:

Use OpenWebUI, n8n, Cline, Cursor with AWS Bedrock and AWS AI services
Access Claude 4.5, Nova, Qwen, DeepSeek, Mistral, Cohere, and 75+ more models
Text-to-speech (Polly), speech-to-text (Transcribe), image generation, embeddings
Keep all data in your AWS account (HIPAA, GDPR, FedRAMP ready)
Multi-region model access from a single endpoint

The Problem We Solved

Most modern AI tools are built for OpenAI's API, but AWS Bedrock uses a completely different SDK. This forces you to choose between using familiar tools or switching to AWS for compliance/cost reasons and rewriting everything.

We ran into this while working with a customer who needed:

Compliance: All data in their AWS account (GDPR requirements)
Choice: Multiple models (Claude, Nova, Qwen, Mistral, DeepSeek, Stability AI, Cohere)
Multi-region access: Different Bedrock models available in different AWS regions
Standard tooling: OpenWebUI, n8n, IDE coding agents (Cline, Cursor)
Full AWS AI stack: Bedrock models plus Polly (TTS) and Transcribe (STT)

Our Approach: An OpenAI-Compatible Translation Layer

stdapi.ai is an API gateway that sits in your AWS account and translates OpenAI API calls to AWS services.

Key Features

OpenAI API Compatibility

Your existing code works without modifications:

from openai import OpenAI

client = OpenAI(
    base_url="https://your-deployment.example.com/v1",
    api_key="your-api-key"
)

# Chat completions
response = client.chat.completions.create(
    model="amazon.nova-pro-v1:0",
    messages=[{"role": "user", "content": "Explain AWS Lambda"}]
)

# Text-to-speech with Polly
response = client.audio.speech.create(
    model="amazon.polly-neural",
    input="Welcome to the future of voice technology!"
)

# Speech-to-text with Transcribe
with open("meeting-recording.mp3", "rb") as audio_file:
    transcription = client.audio.transcriptions.create(
        model="amazon.transcribe",
        file=audio_file,
        response_format="json"
    )

# Image generation with Bedrock
response = client.images.generate(
    model="stability.stable-image-ultra-v1:0",
    prompt="A serene mountain landscape at sunset"
)

# Embeddings with Bedrock
response = client.embeddings.create(
    model="cohere.embed-v4",
    input="Semantic search transforms how we find information"
)

Multi-Region Access

Access models across multiple AWS Bedrock regions from a single endpoint. Different models are available in different regions. You can configure which regions to include based on your requirements. AWS Bedrock inference profiles handle automatic routing when needed.

Multi-Modal Capabilities

Process text, images, videos, and documents together in a single request. Supports HTTP URLs, S3 URLs (for direct access to your data), and base64 data URIs. Perfect for vision tasks, document analysis, and RAG applications.

Deployment

Deploy to your AWS account in 5-15 minutes using Terraform.

Sample configurations set up:

Application Load Balancer with ECS/Fargate
IAM roles with least-privilege access
CloudWatch logging
S3 storage
Optional: Custom domain with ACM certificate

Get started: Sample Terraform configurations

Use Cases

Chat interfaces: OpenWebUI or LibreChat for private ChatGPT alternatives
Workflow automation: n8n connecting AI to 400+ services
Developer tools: IDE coding agents (Cline, Cursor, Continue)
Knowledge management: AI-powered note-taking and semantic search
Internal AI tools: Custom chatbots for Slack, Discord, Teams

Technical Details

How It Works:

Converts OpenAI API format to Bedrock's format
Maps model IDs to appropriate AWS services and regions
Provides unified access to models across configured AWS regions
Converts Bedrock responses back to OpenAI format
Supports streaming via Server-Sent Events (SSE)

Security:

All traffic stays within your AWS account
IAM role-based access control
Optional API key authentication
VPC deployment supported
Integrates with AWS Bedrock guardrails

Performance:

Low latency translation layer
Streaming response support
Scales with ECS/Fargate auto-scaling

Open Source & Commercial Options

stdapi.ai is available as open source (AGPL-3.0) for experimentation and internal use.

For production deployments, we recommend the AWS Marketplace version, which includes:

Hardened container images
Regular security updates
Production support
14-day free trial

The commercial license also removes AGPL obligations for proprietary applications.

Get Started

🏢 Recommended: AWS Marketplace (14-day free trial, production-ready)

🚀 Open source: GitHub (AGPL-3.0 for experimentation)

📦 Sample deployments: Terraform configurations

📚 Documentation: Full guides and API reference

Conclusion

This translation layer approach solves the AWS Bedrock compatibility gap while keeping everything AWS-native.

Key benefits:

Zero code changes: Your OpenAI SDK code works as-is
AWS-native: Fully integrated with Bedrock, Polly, Transcribe
Multi-region: Access models across AWS regions from one endpoint
Production-ready: Terraform deployment, CloudWatch integration

We'd love to hear your feedback! Have you solved this problem differently? What challenges have you faced integrating Bedrock into your workflow?

Vibe Coding Forem