Quarkus Insights #256: What is Floci?

Hector Ventura, a Lead Software Engineer with nearly 18 years of Java experience and eight years working day-to-day with AWS, is the creator of Floci — a free, open-source suite of local cloud emulators built on top of Quarkus native image. In episode 256 he joins the show to walk through what local cloud emulation actually means, give a live tour of the UI, and explain how a part-time side project opened a week before LocalStack archived its community edition became one of the more talked-about tools in the Quarkus ecosystem.

The name Floci is a play on cirrocumulus floccus — the small, popcorn-shaped cloud formation. Hector’s first instinct for the name was "Floccus" directly, but it didn’t feel right, so "Floci" emerged as the working name and stuck. The pronunciation caused some confusion in the episode preview the week before (the hosts attempted roughly six incorrect versions), but it is now firmly settled.

(In case you’re also unsure, saying "flossy" will get you pretty close to what Hector uses.)

Hector draws a clear line between three categories that are often conflated:

Floci is firmly in the emulation category. Although its behaviour is realistic, Hector emphasized that it is not intended to replace a real cloud account in production — its purpose is local development, CI/CD pipelines, and AI agent sandboxing, where fast, isolated, and disposable environments matter far more than billing-grade fidelity.

At launch Floci was a single AWS emulator. Today it covers AWS, Azure, Google Cloud Platform, and Oracle OCI — each running in its own container on its own port, with a consistent CLI and UI experience across all four. The project launched as open source on March 15, a week before LocalStack archived its community edition. Five months in, the volume of incoming issues and pull requests already makes the project feel much older than it is.

For most services Floci implements the cloud provider APIs directly. For services where a from-scratch implementation would take years — databases like AWS RDS being the primary example — Floci uses the same strategy the clouds themselves use: it spins up a real Docker container running PostgreSQL, MySQL, or MariaDB. Lambda is handled similarly: Floci pulls the official AWS Lambda runtime images from the ECR and executes function code inside them, giving near-identical behaviour to a real Lambda invocation.

The entire emulator is compiled to a GraalVM native image via Quarkus, which is what produces the headline numbers the community keeps sharing: 24 ms startup time and 13 MB idle memory.

Quarkus Dev Services originally wired local AWS emulation through LocalStack via Testcontainers. The same mechanism — which talks to the Docker socket to start a pre-configured container — now supports multiple providers, and Floci is one of them. Switching is a configuration change; no application code has to change because your code still talks to the original AWS, Azure, or GCP SDK.

A live demo showed Floci’s built-in status page at localhost:4566, which lists running services and exposes a button to launch the Floci UI as a separate container on port 4500. The UI is deliberately not embedded in the emulator itself — Floci is used heavily in CI/CD and by AI agents, where no UI is needed and keeping the image lean matters.

The UI uses a cloud-agnostic vocabulary (Storage rather than S3, for example) so it can represent all four providers in the same pane. "Coming soon" labels throughout map the intended roadmap in a form far more visible than a GitHub issue list.

For data persistence Floci offers four storage modes:

Hector pointed out that persistence was a first-class design goal from day one, because the LocalStack community edition could not reliably copy or move a persistent dataset — a limitation that drove him to build something different.

One use case that surfaced briefly on screen stood out: the ability to assign a separate Floci instance to each AI agent running in parallel. A company that reached out to Hector described how spinning up isolated Floci environments per agent had materially accelerated their development process. With no real cloud credentials in play and data that simply disappears at the end of a session, it removes the category of horror stories that arise from giving an LLM access to a live account with real billing.

Hector addressed how Floci tracks SDK evolution: AWS SDKs maintain backwards compatibility, so new service versions add capability without breaking existing calls. AWS is working on an interface definition format called Smithy; GCP and OCI are cleaner because they publish gRPC Protobuf definitions that Floci uses directly — the same .proto files the SDKs themselves are generated from.

The demo walked through redirecting the AWS CLI at a local Floci instance. The key points:

Once those three things are in place, aws s3 ls and aws s3 mb work exactly as they would against a real account.

Hector named three priorities:

The release cadence is settling into two releases per emulator per month (first and third Tuesday), down from the early pace of a release every three or four days that was burning maintainer time.

Floci is a young project that arrived at exactly the right moment: a fast, MIT-licensed, Quarkus-native replacement for a tool that tens of thousands of developers depended on. Five months in, the numbers in the issues tracker and pull request queue reflect genuine adoption, and the performance story — anchored in the same native image capabilities that gave this show some of its best benchmark conversations — gives it a strong foundation as it expands from AWS into the full multi-cloud picture. If you are interested in contributing, head to github.com/floci-io to find open issues tagged for new contributors.