AWS Emulator is generally available: a free local AWS for agents and CI

AWS made AWS Emulator generally available on 11 November 2026. It is a program you run on your own machine that answers AWS API calls, so the AWS CLI, the AWS SDKs and coding agents can create and test resources without touching a real account. AWS says there is no additional service fee and that the source is on GitHub.

For a team that has handed infrastructure code to a coding agent, the point is the loop. AWS's announcement says an agent "can create resources, check the result, and iterate again without waiting on a deployment in a Region". It also removes a cost and a risk from that loop: no cloud bill for the agent's trial runs, and no real credentials in its shell.

What you get

• An HTTPS endpoint on 127.0.0.1:9297. You point tools at it with environment variables; AWS says application code stays unchanged.
• Its own credentials, regenerated at each start. Access key IDs begin with EMULATOR and requests signed with any other key are rejected, AWS says.
• Linux packages (RPM and Debian). On macOS and Windows you build a Docker image yourself; AWS says it does not publish a ready-made one.
• An agent quick start. The README installs an aws-emulator skill through the AWS CLI, version 2.35.0 or later, and the announcement names Kiro and Claude Code.
• Embedded mode for Rust tests: one isolated in-process instance per test, so tests can run in parallel without sharing state. The README warns that building it needs at least 100 GB of free disk.
• Broad but uneven coverage. AWS says the emulator is generated from the public AWS API models. The repository's workspace file listed 436 service crates when I read it on 11 November. An operation that is not implemented returns HTTP 501.

The limits AWS states

• It is not a security test. AWS says the emulator does not evaluate IAM policies and resource-based policies the way AWS does, "so a request that AWS denies might succeed in the emulator".
• One account, 000000000000, with no cross-account or cross-Region behaviour.
• Fidelity varies by service. AWS says it publishes, per release, how many fidelity tests pass for each service as k of n, and that the test suite is public. I could not load those scores.
• Telemetry is on by default. The README says usage data goes to AWS without request contents, code or credentials. Turn it off with --telemetry off or EMULATOR_TELEMETRY=off.
• The README says: "Don't use it for production workloads or to test security."

Agents built it, AWS says

AWS says the emulator "is built and maintained by agents" from the API models, running on Amazon Bedrock AgentCore with AWS CodePipeline and AWS CodeCommit. One group of agents compares emulator responses with real AWS and files an issue for each gap. Another picks up an issue, writes tests, implements the behaviour and reviews the change. A separate security agent reviews changes adversarially, and fidelity tests run before anything is published. The repository takes bug reports but, the README says, not pull requests.

What AWS does not give: how many agents, what the build cost, how much human review the code received, or any support or maintenance commitment beyond the word "maintained". The licence is CC0 1.0, with vendored framework code under Apache 2.0.

The pushback on Hacker News

The Hacker News thread on the repository (item 50456631) had about 280 points when I read it. The strongest objection is durability: commenters asked whether an agent-built project will be maintained or was a demonstration. A commenter who writes as part of the team replied that it is "a real product that we plan to maintain"; others answered that a forum comment is not a commitment. One commenter reported roughly 29,000 commits in a month, which I could not verify. Another said there is no quick way to reset state between tests short of restarting the process, which matters for suites built on tools such as moto. Several objected to default-on telemetry.