Claude Platform on AWS: An Architecture Decision Guide for AWS Teams

The Files API lets the application upload and manage files for use with Claude without re-uploading the same content on each request. This is useful for Claude-centered workflows involving uploaded documents, datasets, generated outputs, and code execution.

Bedrock Knowledge Bases are part of the AWS-managed retrieval and customization stack, alongside Bedrock Data Automation.

These are not equivalent features. Claude Files are better for file reuse inside Claude workflows. Bedrock Knowledge Bases are better when managed RAG over enterprise data is part of the architecture.

Claude Platform on AWS supports Claude Managed Agents, Agent Skills, code execution, tool use, credential vaults (rather than full identity management), memory stores, and Claude-specific agent workflows. Skills package instructions, metadata, scripts, templates, and resources that Claude can load when relevant.

Bedrock AgentCore includes features such as Runtime, Gateway, Memory, Identity via IAM, Browser, Code Interpreter, Observability, Evaluations, and Policy, and is designed to work with any framework and model.

Choose Claude Platform on AWS when the agent should be built around Claude’s own agent capabilities and reusable Claude Skills. Choose Bedrock when the agent architecture needs AWS-managed runtime, identity, tool gateway, memory, observability, evaluation, and policy controls.

Claude’s code execution tool runs Python and bash in a sandboxed environment inside the API conversation, enabling calculations, visualizations, file manipulation, and data analysis with uploaded files.

Bedrock AgentCore includes a Code Interpreter as part of a broader AWS agent platform, alongside browser, gateway, memory, identity, policy, and observability capabilities.

Choose Claude Platform on AWS when code execution is part of Claude’s reasoning and file workflow. Choose Bedrock when code execution belongs inside an AWS-managed agent runtime with broader agent controls.

Claude Platform on AWS can be used with application-level controls and Claude-native workflows.

Bedrock Guardrails provides configurable safeguards for inputs and responses, including content filters, prompt attack detection, denied topics, PII handling, contextual grounding, and Automated Reasoning checks. Guardrails works consistently across foundation models and some self-hosted and third-party models.

Choose Bedrock when safety and policy enforcement should be a managed AWS feature applied consistently across applications or models. Choose Claude Platform on AWS when you just need application-level controls.

Claude Console provides Claude-specific prompt generation, prompt improvement, and evaluation workflows. This is useful when teams are tuning a Claude-specific application and want the development loop close to Anthropic’s platform.

Bedrock evaluations can assess models, Knowledge Bases, and RAG sources, including automatic evaluations, judge-model evaluations, and human evaluations. AgentCore recently launched an optimization feature that lets you evaluate and optimize prompts and tool descriptions for agents running on AgentCore.

Choose Claude Platform on AWS when evaluation is centered on Claude prompt and workflow iteration. Choose Bedrock when evaluation needs to compare models, RAG sources, or Knowledge Bases inside the AWS platform, or needs to optimize for tool descriptions taking into account end to end agent metrics.

Message Batches support large asynchronous Claude Messages workloads and can reduce cost for work that does not need immediate responses.

Bedrock also supports batch and cost-optimization patterns, including broader Bedrock pricing and routing features across supported models.

Both options are very similar, this point won't be a deciding factor for you. It is an important cost lever though, so keep this in mind.