Developer open-sources zero-telemetry local AI proxy daemon for data sovereignty

A localhost endpoint that speaks OpenAI

A developer writing on dev.to as JOxKxER has open-sourced the Garza Global Graviton (GGG) Sovereign Edge Daemon, a self-contained background process meant to give local AI workflows a privacy-controlled entry point. According to the post, published on September 19, 2026, the daemon binds to the machine's loopback interface at 127.0.0.1 on port 11834 and serves an OpenAI-compatible API, so tools that already speak the OpenAI request format can be pointed at it without changes.

What the daemon claims to offer

The author's central claim is that nothing leaves the hardware. The post describes the project as built with no accounts, no cloud dependencies and no tracking code, so prompts, conversation history and stored context never move off the user's machine. To support integrity rather than just privacy, the daemon includes what the author calls a "sovereign math framework": responses carry an HMAC-SHA256 tag so tampering can be detected, alongside a "metadata pacer" that the post names but does not explain in detail.

Because the endpoint mimics the standard OpenAI API shape, the project takes a bring-your-own-AI approach. The author lists VS Code's Continue extension, Open WebUI, LM Studio and Jan as compatible tools, arguing that data stays portable because any OpenAI-compatible client can connect. The post demonstrates the setup in Python: the official OpenAI SDK, with the base URL changed to the local endpoint, a placeholder key of "ggg-local" and a model identifier of "ggg-local-sovereign".

There is also an onboarding angle. The repository ships plain-English guidance and copy-pasteable setup prompts, which the author says is intended to let non-technical users secure a local AI pipeline in minutes. The code is hosted on GitHub under the author's account, with a landing page on Render, and the author is asking for feedback and pull requests.

Questions the post leaves open

The announcement is a single self-published post, and the strongest claims are the author's own. A few details would matter to anyone evaluating it.

First, the post does not describe how inference actually happens: whether the daemon bundles or runs a model itself, or forwards requests to another backend. The compatibility list mixes client applications with local inference servers such as LM Studio and Jan, and the intended topology is left unclear.

Second, "absolute data sovereignty", the framing in the post's title, can only extend as far as the daemon itself. If a connected tool or an upstream model provider reaches the internet, prompts can still leave the machine regardless of what the proxy does.

Third, terms like the sovereign math framework and the metadata pacer are asserted rather than documented in the post, so the practical guarantee behind the HMAC branding is hard to assess without reading the source.

Why it matters

Even with those caveats, the project reflects a real shift in how developers approach AI tooling. The OpenAI-compatible endpoint has become a de facto interface, and a local proxy that speaks it can slot into existing editors and chat UIs with a one-line configuration change. That makes "run it locally, keep the data" a practical default rather than a re-architecture.

The demand side is the signal here: developers increasingly want air-gapped options they can audit, and single-maintainer projects like this are where that tooling tends to start. Anyone relying on it for sensitive workloads should treat the privacy claims as a starting point for review rather than a settled fact, but the pattern of small, self-contained, loopback-only services is one worth watching as local AI matures.