Preflight Checklist
Complete this checklist before starting any Get Started tutorial. It ensures your AWS account, model access, and environment are ready so you don't hit permission failures, region mismatches, or missing model access mid-tutorial.
For the full procedural setup, follow the Runtime getting started prerequisites. The sections below cover the planning and awareness context that applies across all paths on this site.
Region
AgentCore services are not available in every AWS region. Before choosing a region:
- Check the AgentCore region availability page for the current list of supported regions.
- Tutorials on this site default to
us-east-1. Substitute a supported region closer to your location if needed. - If you deploy to a region where a required AgentCore service is not yet available, errors will resemble permission failures rather than region availability issues.
Least-privilege IAM permissions
AgentCore tutorials involve two IAM roles:
Deployment role— the identity you use to create AgentCore resources. Scope it tobedrock-agentcore:*on tutorial resources,iam:PassRolefor runtime role association, and any CloudFormation or ECR permissions required by your chosen path.Runtime execution role— the identity AgentCore assumes when running your agent. Scope it tobedrock:InvokeModelfor the specific model your agent uses, plus permissions for any AgentCore services (Memory, Gateway, Observability) and external resources your agent's tools access.
Follow least-privilege from the start. Broad policies like AdministratorAccess on the runtime execution role give the agent unnecessary access to your account. See IAM Permissions for AgentCore Runtime for the permissions required to deploy and run agents.
Bedrock model access
AgentCore is model-agnostic. You can use any LLM provider (Anthropic direct, OpenAI, self-hosted models, etc.) from your agent code. The samples and tutorials on this site use Bedrock for convenience because it runs within your AWS account with no additional API keys. If you bring your own model provider, skip this section entirely.
Access to Amazon Bedrock foundation models is enabled by default in commercial AWS regions when your IAM role has the correct AWS Marketplace permissions. However, there are cases where additional steps are needed:
- Anthropic models require a one-time First Time Use (FTU) form submission per account (or per organization management account). Submit use case details through the Bedrock console or the
PutUseCaseForModelAccessAPI. Access is granted immediately after submission. - Marketplace permissions required — your IAM role needs
aws-marketplace:Subscribe,aws-marketplace:Unsubscribe, andaws-marketplace:ViewSubscriptions. Without these, first invocation of a Marketplace-backed model fails immediately withAccessDeniedException. - Anthropic FTU is checked immediately — if the FTU form hasn't been submitted, invocations fail right away with a
FTUFormNotFillederror, not after a delay. Complete the FTU form before your first Anthropic model invocation to avoid this.
If you encounter an AccessDeniedException referencing a model ID, verify Marketplace permissions and (for Anthropic models) FTU form completion. See Request access to models for the full prerequisites.
Cost awareness
Tutorial workloads typically cost a few dollars or less at development volume. The main cost drivers are:
| Resource | Cost driver | Notes |
|---|---|---|
| Bedrock model invocations | Per-token pricing for input and output tokens | Cost varies by model; larger models are priced higher |
| AgentCore Runtime | $0.0895/vCPU-hour + $0.00945/GB-hour | Scale-to-zero means no cost when idle |
| AgentCore Memory | $0.25 per 1,000 events (short-term); $0.75 per 1,000 records stored/month (long-term, built-in strategies) or $0.25 per 1,000 records stored/month (override/self-managed strategies); $0.50 per 1,000 long-term memory retrieval calls | Minimal at tutorial volume |
| ECR (container images) | Storage per GB/month | Created automatically by agentcore deploy for container-based agents |
| S3 (build artifacts) | Storage and requests | The CLI creates bedrock-agentcore-* buckets for deployment artifacts; negligible at tutorial volume |
| CodeBuild (image builds) | Per build-minute | Used by the CLI to build container images during deployment |
| CloudWatch logs and traces | Ingestion and retention per GB | Usually negligible at tutorial volume |
Resources left running accumulate cost over time. Each tutorial includes a cleanup section — run it before you leave a tutorial behind. See AgentCore pricing for current rates.
Data residency
- Requests to Bedrock models are processed within AWS infrastructure in the region you select. Review the Bedrock data protection page and your organization's data classification policy before processing regulated data.
- If you enable Memory, session context is stored in AgentCore Memory in the deployment region. Understand retention settings before storing anything sensitive.
- Observability traces include prompt content and tool call parameters by default. If a trace contains sensitive data, scope log retention and access accordingly.
Security
Use synthetic or sample data in tutorials — prototype environments lack the access controls required for sensitive data, PII, or proprietary information. Do not substitute production customer data while following tutorials.
For full security guidance:
- Security best practices for AgentCore Runtime — IAM, network, encryption, and session isolation recommendations
- Preparing for production — this site's checklist covering data classification, policy enforcement, and compliance controls
Next steps
- Choose your first agent path if you have not yet selected a tutorial
- Run a code-first prototype with the CLI once your environment is ready
- Prepare for production traffic after completing any tutorial
Everything produced by a Get Started tutorial is a prototype until you have applied the controls in Preparing for production. Do not use a prototype endpoint to serve real user traffic.