AI disclosure
How AI was, and was not, used
Two different questions get asked as one. Whether AI helped write the software is a fact about how it was built; whether the software uses AI on your data is a fact about what it does. They have different answers, and the second one differs by project, so both are answered plainly below.
In development
Parts of the code in these projects, their documentation and this website were written with the help of AI tools — principally Claude. Every AI-assisted contribution is reviewed, tested and maintained by the project maintainer: AI is a productivity aid here, not an unsupervised author. Nothing is committed that has not been read.
That claim is checkable rather than offered on trust, which is the point of the license. The whole result is readable on GitHub (opens in a new tab) — every line, in every project, with the commit history that produced it.
In the projects
There is no blanket promise here that no project will ever use AI, because that is not the commitment. The commitment is narrower and, I think, more useful: AI goes into a product only where it provides actual value, and only as a local model the operator can control and audit. Never a vendor API that your data leaves the building to reach, never a black box you are asked to trust because it is convenient, and never a feature added because the word sells.
The audit part is the load-bearing half. A self-hosted model runs on infrastructure you own, on data that never leaves it, and — because everything here is copyleft with no part held back — the code calling that model is code you can read. An assistant you cannot inspect, running on somebody else's hardware against your logs, fails that test whatever it does for you.
Applied project by project, that comes out as follows.
ihasmail does not use AI to read, process, summarise or generate anything from your mail. It is a JMAP client: messages are held entirely by the Stalwart server its operator runs, ihasmail keeps no database of its own, and no message content, address or attachment is sent to any third-party AI service. There is no such call in the code to make, which is checkable because the source is published. Nothing has been proposed for it that would clear the bar above — a mail client's job is to show you your mail.
Cairn OBS ships an AI feature, and it is the one place in these projects where AI is a product capability rather than a writing aid. It offers AI-assisted query authoring — describe what you are looking for and it drafts the query. Writing a correct query against unfamiliar log structure during an incident is a real cost, and shortening it is real value, which is the test above being met rather than waived. It runs against a self-hosted Ollama model by default: the data stays on infrastructure its operator controls. The assistance compiles down the same query path every hand-written query already runs through, in the same AGPLv3 tree, so what it produced can be read, checked and edited like anything else — and it is not held back for a paid tier. Whether to enable it at all is the operator's call.
stalwart-migrator, SysAdminAutomation and ubuntu2mint use no AI at all. They are a Go binary and two sets of Bash scripts that do a defined job on a machine and exit. stalwart-migrator is standard library only; the scripts use base utilities. None of them makes a network call to anything beyond the task in front of it.
On this site
Nothing you do here is sent to an AI service. These pages run no AI feature of any kind: no assistant, no chat, no summarising, no scoring or profiling of visitors. Analytics, if you accept it, is Google Analytics as described in the privacy notice and nothing more.
Questions
Ask on GitHub issues (opens in a new tab), or email info@linuxexperts.net. If something on this page is out of date because a project changed, that is worth telling me about.