Trust & security
Releezy exists so your team can release with trust. That only works if you can trust Releezy itself: with your data, with your isolation, and with a measurement that never bends, not even for our own agents.
This page explains, plainly, what we access, what we never touch, and why the ruler that measures your team is the same one that measures us.
The four pillars
Everything on this page comes down to one commitment: what you see on the scoreboard is true, and what belongs to you stays yours.
01
Releezy Guardian reads only what the measurement needs: pull request history, review comments, and change metadata, read-only, straight from your git repository. Your source code is never sold, never shared, and never used to train anything.
02
Every customer operates in its own isolated environment, with its own data, its own agent runs, and no shared state. When Releezy Loop executes an agent, it works inside a dedicated, disposable environment that is discarded when the run ends. Nothing crosses the line between customers.
03
Guardian's standard is fixed. Its metrics are not tunable to flatter Releezy's own agents, and every run our modules make lands on the scoreboard, including the bad ones. If a Releezy module underperforms your team, you see the number before you see our marketing.
04
The baseline is never a vendor benchmark. Your best human reviewers, on your own codebase, define what good looks like, and every contributor, human or agent, is measured against them. Including ours.
The standard
A company that ships both the measurement system and the agents being measured faces an obvious temptation: soften the measurement. Re-weight the metric. Redraw the category until the score looks kinder.
We made the opposite commitment, and we gave it a name: the direction of adaptation. In plain terms, it means that when one of our agents underperforms your human baseline, the agent is what changes. The standard never moves. Not for Releezy Loop, not for Releezy Reviewer, not for anyone.
That is why you can put your humans and our agents on the same scoreboard and trust the result. The ruler is the fixed point. Everything else, including us, adapts to it.
Integrity and integration can coexist. The fixed standard is what makes it possible.
Pull request history, review comments, and change metadata, read-only. That is what the measurement needs, and that is where the access ends. Your source code is never sold, shared, or used to train anything, and you can revoke access at any time.
The standard is fixed by design. Guardian's metrics are not tunable per contributor, so there is no dial we could turn in our own favor. Every run by a Releezy agent lands on the same scoreboard as your team, including the runs we would rather not show you. When one of our agents underperforms, the agent changes, never the measurement.
Yes. Each customer's data lives in its own isolated environment, and agent runs execute inside dedicated, disposable environments that are discarded when the run ends. Nothing is shared across customers: not data, not context, not results.
A 30-minute demo runs on your pull request history and shows you the same scoreboard we hold ourselves to. Trust built with data, starting with yours.