Archibald312 lays the foundations for a hard fork

A wave of engineering plumbing makes GordonOSS ready to live on its own - and quietly draws a privacy line that matters to anyone routing client documents through AI.

infrastructuresecurity

Archibald312 spent this round on the unglamorous but essential work of standing a fork up as its own project: automated checks that run on every code change, a suite of tests covering login, access controls, and how stored API keys are encrypted, plus end-to-end tests that walk through real user flows like uploading documents and using chat. A FORK.md spells out the relationship to upstream, including a firm don't-send-changes-back warning and a nod to the open-source licence the project inherits.

The most consequential change isn't infrastructure though - it's a guard that blocks customer documents from being routed through free tiers of Google's Gemini AI models unless an administrator explicitly opts in. Free-tier model traffic can be used to train the underlying AI, so the default now treats that as a decision, not an accident.

So what Worth a glance for legal-ops leads evaluating AI tools: this is what 'we won't quietly leak your documents into training data' looks like in code.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?