Aigis as a tool for your AI assistant

This case study is real; every step in it is a call that the Aigis interface offers today.

A mid-sized machine builder wants to outsource the maintenance software for its production lines. A supplier has sent its specification, 64 pages. Against it stand the company's own IT security policy, the group's data protection requirements and the requirements from the NIS 2 implementation. The question is easy to ask and hard to answer: what of this does the specification meet, what not — and where do the company's own requirements contradict each other?

The team works in its development environment with GitHub Copilot in agent mode. That is where it stays. Aigis is not brought in as one more interface, but as a tool of the assistant.

The connection takes three steps:

• In Aigis, the project lead creates a key under "API & Agents". It is personal to her and can do exactly what she may do in Aigis — no more.
• She stores the key as an environment variable, not in the chat. The assistant reads it from there.
• She gives the assistant the address of the operating manual: aigis.one/api/llm-guide?module=project. This manual is written for AI agents, not for people. It describes every call, every limit and the rules — for example, that every step is released individually.

After that it is sentences in plain language:
"Read the Aigis manual for projects. Create the project 'Maintenance Line 3'. The specification goes into the specification rubric; security policy, data protection requirements and NIS 2 requirements go into the requirements rubrics. Tell me what the evaluation will cost before you start it."

The assistant creates project, rubrics and documents in a single call, asks for the cost and waits. After the release it starts the evaluation of the rubrics, then the GRC check across everything. It polls the status at intervals instead of pressing, and reports back when the result is ready.

In less than two hours it is on the table. Assumed for this example:

• eleven requirements of the security policy that the specification does not cover, among them the deadline for security updates and the logging of remote access;
• two places where the specification contradicts the data protection requirements — storage location of the maintenance logs and deletion period;
• one contradiction in the company's own requirements: the security policy demands a retention of access logs that the data protection requirement does not permit.

Every point carries both citations: document, section, wording.

Then the part begins that the assistant is there for anyway:
"Export the result as Markdown. From it, draft a request for additions to the supplier, one paragraph per missing requirement with the citation. Do not send the contradiction in our own requirements to the supplier; write it as a question to the data protection officer and the head of IT security."

Aigis delivers the finding, the assistant turns it into the day's work: the letter to the supplier, the internal query, the task list in the ticket system. No retyping, no copying from PDF pages, no citation that gets lost on the way.

What it brought, in a calculation with explicitly assumed figures: without Aigis, two specialists — information security and data protection — read the specification against four sets of rules. Let us assume three working days each, so six person-days, and the result is a table whose citations one has to take on trust. With Aigis and the assistant, the same is ready in less than two hours, with evidence, and the two specialists spend their time on what only they can do: deciding which side of a contradiction applies, and negotiating with the supplier. The price of the evaluation hardly carries weight in this calculation; the week by which the award would otherwise have slipped weighs more.

Two things Aigis does not do here. It does not decide which requirement applies — it discloses the contradiction with both sides. And the assistant may do no more than the person it acts for: the key carries that person's rights, and every action is in the log under that person's name.

So whoever works with an AI assistant does not have to learn a new interface for Aigis. One key, one address, one sentence — and Aigis is one of the assistant's tools.