AI agents · OpenClaw · self-hosting · automation

Quick Answer

AWS Emulator vs LocalStack vs Moto: Local AWS Testing 2026

Published:

The short answer

AWS Emulator is the new default for local AWS testing; LocalStack is what you pay for when you need IAM, state and support; Moto stays the fastest choice inside Python unit tests. As of November 15, 2026:

AWS EmulatorLocalStack for AWSMoto
MakerAWS (built and maintained by AI agents)LocalStack GmbHOpen-source community
Released / statusGA November 11, 2026; v1.0.385 on November 14, 2026Unified image since March 20265.2.5 (October 27, 2026)
PriceFree, no service feeHobby free (non-commercial); Base $39/licence/month annual ($45 monthly); Ultimate $89; Enterprise customFree
LicenceCC0-1.0Commercial (account + auth token required)Apache-2.0
Language / runtimeRust; local HTTPS endpoint on 127.0.0.1:9297Docker containerPython library; optional moto_server
Service coverageEvery service with a published AWS API model; fidelity scored per service30+ (Hobby), 55+ (Base), 110+ (Ultimate)Many services, mocked per API
IAM policy evaluationNoBasic (Base), advanced (Ultimate)Limited
State persistenceIn memory; embedded mode isolates state per testLocal + cloud state on Base and aboveIn memory
Best forCLI/SDK tests and coding-agent loopsTeams needing IAM checks, persistence, CI at scale, supportPython unit tests

What AWS Emulator is

AWS announced AWS Emulator on November 11, 2026 as a local emulator that the AWS CLI, AWS SDKs and coding agents reach over a local HTTPS endpoint using the same APIs as the cloud. Because it is generated from the same API models the SDKs use, request and error shapes match AWS. On start it writes credentials, a config file and a certificate to ~/.aws/emulator/9297/; you point tools at it with AWS_ENDPOINT_URL=https://127.0.0.1:9297 and the emulator’s credentials file. Its access keys start with EMULATOR, and it rejects requests signed with any other keys — a useful guard against a test accidentally hitting a real account.

Two details matter for test suites:

  • Embedded mode runs the emulator inside the test process with its own state per instance, so tests can run in parallel without sharing buckets or tables. It is available only for the AWS SDK for Rust at launch.
  • Fidelity scores: AWS publishes, per release, how many of each service’s fidelity tests pass (“k of n”), and the test suite is public. Unimplemented operations return HTTP 501 so tests fail fast instead of silently passing.

AWS is explicit about the limits: no IAM or resource-policy evaluation, single account, no cross-Region behaviour, and anonymous usage telemetry on by default (it can be turned off). Do not use it to test whether your application is secure.

Where LocalStack still wins

LocalStack has the features AWS Emulator lacks: IAM policy enforcement, persistent state you can save and share (Cloud Pods), application preview sandboxes, an AWS resource replicator on Ultimate, chaos/resiliency testing, Kubernetes delivery and air-gapped images on Enterprise. If your CI checks that a role cannot read a bucket, AWS Emulator will give you a false pass and LocalStack will not. The trade-off since March 2026 is that every user needs an account and auth token, and commercial use needs a paid licence.

Where Moto still wins

Moto patches boto3 inside the Python process with a decorator, so there is no server to start and tests run in milliseconds. For Python unit tests that touch a handful of services it remains the lowest-friction option, and moto_server covers other SDKs if you need it. It is not designed for full integration tests or for agents that drive the AWS CLI.

Picks by situation

SituationPick
Claude Code, Codex or Kiro iterating on AWS code locallyAWS Emulator
Python unit tests with boto3Moto
CI must verify IAM denials or cross-account accessLocalStack Base/Ultimate, then a real AWS test account
Rust service with parallel testsAWS Emulator embedded mode
Air-gapped or regulated enterpriseLocalStack Enterprise
Hobby project, zero budgetAWS Emulator (LocalStack Hobby is non-commercial only)

Whichever you choose, keep a final stage that runs against a real, throwaway AWS account: all three tools tell you about API shape, none proves production behaviour. For sandboxing the agents themselves, see best AI agent sandboxes and AWS MCP Server vs Claude MCP vs third-party.

Last verified: November 15, 2026.

Sources