Why this exists

An AI that can read your servers is genuinely useful. An AI that can change them is genuinely dangerous. Most tools in this space pick one and hope you don’t notice the other: they either stay read-only and stop being useful the moment something needs fixing, or they hand a model a shell and call the risk “autonomy”.

Neither is a real answer for a machine that serves your customers at three in the morning.

Noctrel’s bet is that the useful part and the dangerous part can be separated, and that the boundary belongs in code rather than in a prompt.

So the model gets to reason freely — read live state, form a diagnosis, decide what it would do. It does not get to act freely. Between the two sits a catalogue you wrote, a risk classifier that pattern-matches command text, and a human who clicks Approve.

The distinction matters because prompts can be talked around and code cannot. A model asked nicely not to run rm -rf may still do it. A model that can only ever propose entries from a list you wrote cannot.

One developer — Emirhan Cabuk. Noctrel is not a company. There is no team page to put here, no investors, and no headcount.

That is a real trade-off and worth saying out loud rather than hiding behind a plural “we”: it means slower, and it means the bus factor is one. What it buys is that nothing ships because a roadmap said so, and the person who answers your email is the person who wrote the line of code you are asking about.

No support tier, no account manager, no sales process.

Self-serve — create an account and pair a server yourself, no waitlist. It works end to end — an agent pairs, the AI proposes, a human approves, the command runs, the result comes back and the audit log records all of it — and the approval, risk-classification and execution paths are covered by tests that fail the build if they regress.

It is still early software. The terms say so bluntly rather than burying it.