bedrock_skill

Overview

This skill provides an interface to AWS Bedrock foundation models for building generative AI applications. It exposes patterns for text generation, streaming responses, embeddings, conversation state, model listing, and model access management. Use it to integrate Claude, Titan, Llama, Mistral, and Stable Diffusion models through a unified API and boto3/CLI examples.

How this skill works

The skill wraps Bedrock Runtime and control-plane operations to invoke models synchronously or with streaming, create embeddings, and manage multi-turn conversations with history. It demonstrates common CLI and boto3 calls, handling model IDs, content types, and request/response JSON bodies. It also covers model discovery, access requests, and error handling patterns like retries and exponential backoff.

When to use it

• Building text-generation features (summaries, Q&A, code generation).
• Creating embeddings for retrieval-augmented generation (RAG) pipelines.
• Streaming long-form outputs to improve user experience.
• Managing model access, listing available models, or requesting EULA acceptance.
• Optimizing inference cost and throughput for production workloads.

Best practices

• Choose the smallest model that meets quality needs to reduce cost and latency.
• Set max_tokens and cache repeated responses to limit spend.
• Use streaming for long responses and provisioned throughput for high-volume steady traffic.
• Apply least-privilege IAM, enable VPC endpoints, and audit invocations with CloudTrail.
• Implement exponential backoff and retries for ModelNotReady and throttling errors.

Example use cases

• Invoke Claude to generate technical explanations or developer docs.
• Produce 1024-dimension embeddings with Titan for semantic search and RAG.
• Stream model output to a chat UI for progressive display of long answers.
• Maintain conversation history for a multi-turn assistant with system prompts.
• Automate model discovery and access checks before running inference jobs.