AWS AI services — Bedrock (Claude, Llama, Titan), SageMaker, Rekognition, Comprehend, Transcribe.
Stub — full skill pending.
| Service | Purpose |
|---|---|
| Bedrock | Managed foundation model API — Claude (Anthropic), Llama (Meta), Titan (Amazon), Mistral |
| SageMaker | End-to-end ML platform — training, fine-tuning, hosting, MLOps |
| Rekognition | Computer vision — object detection, face analysis, OCR, content moderation |
| Comprehend | NLP — entity extraction, sentiment, key phrases, classification |
| Transcribe | Speech-to-text; real-time and batch |
| Polly | Text-to-speech |
| Textract | Document extraction — forms, tables, handwriting from PDFs/images |
import { BedrockRuntimeClient, InvokeModelCommand } from '@aws-sdk/client-bedrock-runtime';
const client = new BedrockRuntimeClient({ region: 'us-east-1' }); // Bedrock not yet in af-south-1
const response = await client.send(new InvokeModelCommand({
modelId: 'anthropic.claude-opus-4-6',
contentType: 'application/json',
accept: 'application/json',
body: JSON.stringify({
anthropic_version: 'bedrock-2023-05-31',
max_tokens: 1024,
messages: [{ role: 'user', content: prompt }],
}),
}));
const result = JSON.parse(new TextDecoder().decode(response.body));
const text = result.content[0].text;Note: Bedrock is not available in af-south-1 as of 2025. Use us-east-1 or eu-west-1 for Bedrock. Document the cross-region transfer for POPIA purposes if processing personal data.
Prefer Cloudflare AI Gateway → Anthropic API directly for most 2nth.ai use cases — lower latency, no AWS SDK dependency, works from Cloudflare Workers natively.