All Resources
One place to search every blog, video, code sample, and guide published for Amazon Bedrock.
Introducing the Claude apps gateway for AWS
A self-hosted control plane for running Claude Code and Claude Desktop across an organization without per-developer cloud credentials. The gateway centralizes identity via OIDC, policy enforcement, OpenTelemetry telemetry, request routing to Amazon Bedrock or the Claude Platform on AWS, and daily/weekly/monthly spend caps per org, group, or user, deployed as a stateless container on ECS, EKS, or EC2 behind an ALB with Amazon RDS for PostgreSQL.
Claude Apps Gateway
A self-hosted proxy for running Claude Code and Claude Desktop on AWS with centralized control — OIDC/SSO authentication, team-based model and tool policies, per-user spend caps, OpenTelemetry cost tracking, and multi-region failover routing, so developers never hold individual AWS credentials or API keys.
Automated Reasoning checks in Amazon Bedrock Guardrails add new policy refinement workflows
Automated Reasoning checks in Amazon Bedrock Guardrails add two new policy refinement workflows, an iterative improvement workflow and an ambiguity reduction workflow, to strengthen policies with less manual work.
Build a local LiteLLM AI gateway to test model routing and cost attribution
A fully local LiteLLM proxy demo that fronts multiple providers behind logical model groups, exercising load balancing across Groq Llama 3.1 8B and Amazon Nova Micro, and fallback from Gemini 2.5 Flash to Claude Haiku on Bedrock. Shows how a gateway centralizes routing, retries, request logging, and per-user cost attribution so adding or swapping a model is a config edit rather than a code change.
Embed the world: Multimodal AI for searchable aerial imagery at scale
How the AWS Generative AI Innovation Center and aerial imagery provider Vexcel built a system that makes large-scale aerial imagery searchable via natural language. The approach combines multimodal embeddings, LLM-generated captions, and vector search, benchmarked across roughly 100 configurations against OpenStreetMap ground truth, with caption integration delivering the biggest accuracy gains.
Reliable AI Assistants for Chip Design on AWS
Resilience patterns for keeping a self-hosted, LLM-powered chip design assistant available, accurate, and secure under tape-out load. The assistant runs on Amazon EKS inside the customer VPC, uses Amazon Bedrock via VPC endpoints for inference and a RAG layer backed by Amazon Bedrock Knowledge Bases, and fronts every model call with an LLM gateway that adds multi-model routing, circuit breakers, semantic caching, retries with fallback, and audit logging.
Give Every Knowledge Worker a Research Assistant That Never Sleeps: Claude Cowork on Amazon Bedrock
An overview of Claude Cowork, a macOS and Windows desktop application that delegates research, document analysis, report generation, and workflow automation to Claude running exclusively through Amazon Bedrock in your AWS account. Covers the architecture (InvokeModel calls authenticated with IAM, no intermediate server, CloudTrail and VPC endpoints), the security posture, and per-token pricing for the available models.
One Platform, Every AI Workload: A Practical Guide to Choosing the Right Claude Model on Amazon Bedrock
A decision framework for picking among the Claude models available on Amazon Bedrock, with a model-family table covering context windows, per-token pricing, and best-fit workloads. Maps each tier to tasks from high-volume classification on Claude Haiku 4.5 through daily coding on Claude Sonnet 4.6 and agentic coding on Claude Opus 4.7.
RAG Was Painful Until Now: How Amazon Bedrock Managed Knowledge Base Eliminates the Plumbing
A walkthrough of Amazon Bedrock Managed Knowledge Base, the fully managed RAG service announced GA at AWS Summit NYC, which collapses storage, retrieval, embeddings, re-ranking, and model selection into one managed primitive. Covers its three headline capabilities: native data connectors for sources like Amazon S3, SharePoint, and Confluence; Smart Parsing that preserves tables and document structure; and an Agentic Retriever for multi-step queries.
The Token Efficiency Playbook: 10 Methods to Spend Less on LLM Inference
A practical survey of 10 token-reduction techniques for LLM inference, from zero-risk API prompt caching to research-frontier KV-cache and prompt compression, with peak gain, quality impact, and production-readiness for each. Provider-agnostic; references Anthropic, OpenAI, and academic work, and notes prompt caching makes cached input tokens roughly 10x cheaper.
Building a chat-based Marp slide generator with Amazon Bedrock and Lambda containers
Walkthrough of an app that turns natural language into Marp slides through a chat interface and exports to PDF and editable PPTX. Uses the Amazon Bedrock Converse API with the us.anthropic.claude-sonnet-4-6 cross-region inference profile for multi-turn editing, an AWS Lambda container image colocating Chromium and LibreOffice, Amazon S3 for output delivery, and keyless AWS access from Vercel via OIDC federation. Also covers a multi-stage Docker build that shrinks the Lambda container by 25%.
Cross-Region Private Connectivity for Bedrock Models
An architectural pattern using cross-region AWS PrivateLink to privately route traffic from a workload's region to Amazon Bedrock endpoints in another region, keeping all data on the AWS backbone. Covers the architecture, data flow, security considerations, and trade-offs.
Implementing an emergency stop control for Amazon Bedrock model invocations
A production-ready, layered approach to instantly halt and cleanly restore Amazon Bedrock model invocations on demand. Combines an AWS AppConfig application feature flag for graceful control with an IAM deny policy as the enforced hard stop, plus automatic triggers from CloudWatch token-spike alarms and AWS Budgets cost thresholds converging on a Lambda toggle function. Includes the implementation pattern and a validated walkthrough.
sample-code-for-ace-step-music-generation-on-sagemaker
A sample that deploys ACE-Step v1.5 on Amazon SageMaker AI for asynchronous music generation, with lyric generation via Amazon Bedrock (Anthropic Claude Haiku 4.5) and a static front end on Amazon S3 and CloudFront. Three Lambda functions behind an API Gateway HTTP API handle music generation, lyric generation, and song retrieval, all provisioned by a single AWS SAM template.
Claude Cowork on Amazon Bedrock with JumpCloud Integration
How to set up Claude Cowork to use Claude models hosted in Amazon Bedrock while using JumpCloud and AWS IAM Identity Center for single sign-on, from the perspective of an organization administrator.
From Raw Data to AI Insights: Building a Serverless Analytics + GenAI Dashboard with Athena & Bedrock
A fully serverless analytics system that combines Amazon QuickSight dashboards with a two-step Amazon Bedrock pipeline (intent extraction plus data-grounded explanation, using Nova Lite) to answer natural language questions over manufacturing data queried through Amazon Athena.
How I Built a Fully Serverless RAG Chatbot on Amazon Bedrock for My University
A step-by-step walkthrough of building SUChatBot, a serverless RAG chatbot using Amazon Bedrock Knowledge Bases, S3 Vectors, Claude Haiku 4.5, and AWS Lambda, with the architectural tradeoffs behind every decision.
sample-privacy-preserving-document-ai-bedrock
A pipeline that runs AI analysis on financial documents without exposing personal information to the large language model. PII is detected and masked first, Amazon Bedrock Guardrails verify the masking, the model analyzes only anonymized text on Amazon Bedrock (Claude Sonnet 4.6), and original values are restored deterministically in the final report. The token-to-original map is stored in DynamoDB with a short TTL, and sample inputs are provided in Korean and English.
Amazon Bedrock Managed Knowledge Base is now generally available
Amazon Bedrock Managed Knowledge Base is now generally available as a fully managed RAG service for building production-ready AI agents grounded in enterprise data, with six native data source connectors and AgentCore integration.
Bedrock Disaster Recovery: Building Resilient Applications
Amazon Bedrock is managed, but application-level resilience is your responsibility. This post walks through fault tolerance patterns for Bedrock applications: the shared responsibility model, state management, replay capabilities, context window guardrails, quota-aware routing, and the foundation-model gateway pattern that ties them together.
Building a Multi-Agent Trading Intelligence System on AWS with Strands Agents SDK, Amazon Bedrock, Aurora PostgreSQL, and DynamoDB
How TradeWizard orchestrates 14 specialized AI agents across 7 foundation models using the Strands Agents SDK and Amazon Bedrock, running serverless in AWS Lambda. Aurora PostgreSQL Serverless v2 acts as the durable brain for agent memory and performance scoring, with DynamoDB as the speed layer for dedup cache, WebSocket connections, and rate limiting.
My AI Agent Journey: From Local Ollama to an MCP-Powered AWS Architect
A step-by-step walkthrough of an AWS Builders Skill Sprint, progressing from a local Ollama agent to a cloud-connected AWS Documentation Chatbot built with the Strands SDK, Amazon Nova Pro on Amazon Bedrock, and the Model Context Protocol (MCP). Covers adding custom tools, mem0 memory, and debugging the MCP documentation server integration.
Safeguard your agentic AI applications with the Amazon Bedrock Guardrails InvokeGuardrailChecks API
Announces the Amazon Bedrock Guardrails InvokeGuardrailChecks API, a resourceless, detect-only API that applies individual safeguards at any point in an agentic AI workflow without provisioning guardrail resources. It returns numeric scores per safeguard so applications can set custom thresholds and actions (block, retry, escalate, or log), and contrasts the new API with the existing ApplyGuardrail API.
Unlocking Bedrock Cost Allocation
An episode of "The Keys to AWS Optimization" on allocating and optimizing Amazon Bedrock costs. Covers token-based Bedrock pricing, the new ability to tag IAM roles and session data for cost allocation, enabling IAM principal allocation in cost and usage reports, visualizing spend with the Kudos dashboard, model lifecycle tracking, and tagging best practices.
How do I troubleshoot the "503 Service Unavailable" error in Amazon Bedrock?
An AWS Knowledge Center video walking through how to troubleshoot the "503 Service Unavailable" error in Amazon Bedrock.
Running Claude Code on Amazon Bedrock
A complete walkthrough of running Anthropic's Claude Code on Amazon Bedrock so data stays within AWS, covering environment setup, cross-region inference, prompt caching, the Amazon Bedrock Mantle endpoint, per-team cost tracking, and Amazon Bedrock Guardrails for content governance.
Amazon Bedrock Guardrails adds automated reasoning checks in Sydney
Amazon Bedrock Guardrails Automated Reasoning checks, which use formal verification to validate model outputs and detect hallucinations, are now available in the Asia Pacific (Sydney) Region.
Amazon Bedrock Guardrails announces a new API targeting agentic AI workflows
Amazon Bedrock Guardrails introduces the InvokeGuardrailChecks API, a resourceless tool to apply individual safety checks such as content filters, prompt attack detection, and PII filters at any step of agentic AI workflows, with per-request control and numeric severity and confidence scores.
Introducing Gemma 4 models on Amazon Bedrock
Announces the availability of Google DeepMind's Gemma 4 family of open-weight models on Amazon Bedrock, accessible through the OpenAI-compatible bedrock-mantle endpoint. The three instruction-tuned variants support built-in reasoning, native function calling, and multimodal text and image input across dense and mixture-of-experts architectures, with a walkthrough of console usage, the OpenAI SDK, service tiers, and on-demand inference scaling.
The Future is Conversational: Analyzing Cloud Networks with GenAI
A recap of a talk from AWS Student Community Day Tirupati on using Amazon Bedrock to analyze VPC Flow Logs conversationally. Instead of building complex SQL or Athena queries over logs stored in Amazon S3 and CloudWatch, the approach pre-processes log data in Python, builds a structured prompt, and streams it to Amazon Bedrock so engineers can ask natural-language questions about network activity.
sample-genai-app-builder-with-amazon-bedrock
An AI-powered web development agent that lets you prompt, run, edit, and deploy full-stack applications directly from the browser, powered by Amazon Bedrock foundation models and agentic capabilities. It generates HTML, CSS, JavaScript, and React components, supports iterative refinement and brand templates, and runs generated code in an isolated ECS sandbox container. Deployed via AWS CDK with Cognito auth and a CloudFront front end.
Building a Fault-Tolerant AI Gateway for Amazon Bedrock
How circuit breakers, chaos engineering, and graceful degradation keep AI applications available when foundation models fail. Walks through LLM-specific failure modes (throttling, latency degradation, model deprecation, quota exhaustion, silent quality drops, regional outages) and resilience patterns for a custom LLM gateway built on Amazon EKS with Amazon Bedrock, including circuit-breaker multi-model fallback and cross-region inference.
Building AI Agents Using Amazon Nova Pro and Strands SDK
A walkthrough of building AI agents with Amazon Nova Pro and the Strands SDK across five hands-on challenges, covering tool use, agent memory, and Model Context Protocol (MCP) integration.
Building Supercharger: How Rocket Close optimized title operations with agentic AI
Describes how Rocket Close, a title agency within Rocket Companies, partnered with AWS to build Supercharger, an agentic AI solution that streamlines title operations workflows. The system uses natural-language conversation to centralize knowledge and automate research-heavy, state-specific title examination tasks for operations teams. The article covers the solution's architecture, technology choices, lessons learned, and business impact.
From Image to Decision: Solving the Inspection-and-Report Problem at Scale with NVIDIA Nemotron models on Amazon Bedrock
A solution pattern that uses a single NVIDIA Nemotron pipeline on Amazon Bedrock to turn images into review-ready reports across industries such as manufacturing quality control and healthcare diagnostic imaging, pairing vision-language and reasoning models with human-in-the-loop review at serverless scale.
Stop Guessing on Throttles: Mapping Amazon Bedrock Error Codes to the Right Service Tier
A decision matrix mapping Amazon Bedrock runtime errors (429s, 503s, timeouts, and 4xx client errors) to their root causes and the service tier or fix that actually resolves each, covering the Reserved, Priority, Standard, and Flex tiers and when to use Provisioned Throughput or cross-region inference.