AivexaNewsSearch
AI news for builders and product teamsChecked every hour

AWS Open-Sources Emulator for Local Testing of 200+ Services by Coding Agents

Collected Nov 16, 2026

AWS has open-sourced AWS Emulator, a local program that exposes an HTTPS endpoint standing in for the public APIs of more than 200 AWS services, so developers and coding agents can create resources and reproduce AWS-style errors without provisioning real cloud infrastructure or paying for the emulated operations. The project is generally available at no additional service fee, and the source is on GitHub under the CC0 1.0 public domain dedication, except third-party code in the vendor/ directory, which stays under Apache License 2.0.

Pointing existing tooling at the emulator takes environment variables such as AWS_ENDPOINT_URL for the AWS CLI or an SDK, so application code does not need to change. Installation differs by platform. On Linux it ships as an RPM or Debian package for x86_64 or aarch64. On macOS and Windows AWS documents a short Dockerfile rather than publishing a prebuilt image. Rust tests can use an embedded mode that runs the emulator inside the test process and wires its HTTP client into the AWS SDK for Rust. An aws-emulator skill can launch the emulator for coding agents and check a project against the published fidelity tests.

The software is written in Rust and generated from more than 400 Smithy API models that AWS publishes for its SDKs, according to Swaminathan Sivasubramanian. Smithy is AWS's own interface definition language for describing service APIs; generating the emulator from those models means its surface area follows the same contracts the real SDKs are built from, at least at the shape-of-the-API level.

AWS says agents wrote and tested the code. Matthew Goldberg, a senior principal engineer at AWS, describes three teams: one writes fidelity tests from the public API models and verifies them against real AWS, one replays those tests and files issues, and one implements each fix. Every change must pass a merge gate checking compilation, coding standards, security and fidelity regressions. The work runs on Amazon Bedrock AgentCore Runtime Instances, AWS CodePipeline and AWS CodeCommit.

Accuracy is uneven across the catalog, and AWS is explicit about it. Versioned fidelity tests and per-service scores ship with each release, and a test counts as passing only when every call matches real AWS, including status codes, side effects and exact error messages. Unimplemented operations return HTTP 501 with the error code NotImplemented. AWS says fidelity is highest on control-plane APIs, while data-plane behavior and support for CloudFormation, Terraform and the AWS CDK vary by service. The published evaluation suite also includes tests the emulator does not yet pass.

Documented limits matter for anyone planning to rely on it. IAM and resource-based policies are not evaluated, so the tool must not be used for security testing. Each process serves a single shared account and Region. State lives in memory and vanishes when the process stops. Long-running and load testing are out of scope.

Outside commentary has focused on the test numbers and packaging. Developer Josh Builds counted the published fidelity tests and reported that S3, DynamoDB and SQS pass roughly two in three, while CloudFormation passes under one in five. Yurii Serhiichuk compared the tool with existing emulators and published a container image and GitHub Action to fill a packaging gap, describing it as similar in spirit to LocalStack or Floci but native to AWS.

The emulator enters a field already served by LocalStack, Floci and moto. Its differentiators, based on the stated facts, are that it is generated from AWS's own published API models, carries official versioned fidelity scores and is placed in the public domain. In practice those scores and the missing Docker image and GitHub Action suggest it is not yet a drop-in replacement for every workflow the older tools cover, particularly permission testing and infrastructure-as-code pipelines. The 501 response for unimplemented operations is a useful design choice: failures are loud rather than silently simulated, which helps coding agents tell the difference between "the emulator cannot do this" and "my code is wrong."

Developers get no cost alongside fidelity. Testing against a local endpoint removes the latency, credentials and cleanup overhead of real accounts, which is exactly the loop agent-driven code generation needs: write, run, read errors, fix. The in-memory, single-account, single-Region model simplifies setup at the price of any test that depends on persistence, cross-account access or multi-Region behavior. Because policies are not evaluated, teams will still need real AWS or another tool to validate permissions. And the per-service scores mean a project touching many services will hit uneven coverage, with infrastructure-as-code paths appearing weakest in the reported counts. Setup details are in the user guide, and fidelity scores ship with each release.

Why it matters: teams building with coding agents now have a first-party, no-account, no-cost endpoint for iterating against more than 200 AWS APIs, with published per-service fidelity numbers to decide where it is trustworthy. My inference is that the loud 501 fallbacks and official scores make it a reasonable default for control-plane unit tests and agent feedback loops, while LocalStack, moto or real accounts remain necessary for IAM, load and infrastructure-as-code work until coverage improves. The public-domain license and the absence of an official container image also lower and raise adoption friction respectively; third-party images and Actions already exist to close that gap.

Read at InfoQ · AI, ML & Data Engineering

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

AWS Emulator is a generally available local HTTPS endpoint, generated from public Smithy API models, that lets developers and coding agents test against more than 200 AWS services without an account, with published fidelity scores and documented limits around IAM, regions, and in-memory state. By Daniel Dominguez