Questions answered from the source
Staff, technicians and customs officers get the answer from your own policies, publications and guidance, with the paragraph or the page cited.
Defence and public safety · AI agents inside the organisation
Desk work only. An officer decides. Data stays.
Katonic runs AI agents inside your own data centre, a sovereign cloud or a network with no outside connection. Agents read your policies, your orders and publications and your records, prepare the paperwork, and stop for a named officer or official.
No agent takes part in targeting or any use of force, identifies, profiles or scores a person, or decides to arrest, charge, detain, search, admit at a border or dispatch. It prepares paperwork, and a named officer decides.
Records, case files and personnel data stay on infrastructure you control.
Records, case files and personnel data stay on your infrastructure
A model you serve
Each person's permissions and markings kept
No agent decides about a person
Policy questions, maintenance records, bids, disclosure requests and reports run on reading, checking and drafting. That is where an agent helps, and where a wrong line in a record or a report costs most.
Air-gapped offers exist from several large suppliers in this sector. Running disconnected is the starting point, not the answer: what matters is which tool waits for a person, whose identity the agent uses, and what is recorded.
The police report-drafting products we read are vendor cloud services.
One published case shows an assistant's output used in a police report with no human check. Here a report drafted with an agent is marked as such, the notes are kept, and the officer signs and owns it.
Tools must be chosen one by one: of 27 public interface descriptions at the edge of this work (procurement, customs and border, tenders and sanctions lists), the importer refused none; they carry 123 operations, 60 sent as changes. The operational systems publish no public description, so the organisation supplies its own inside its network.
Personnel records, case files and equipment records are sensitive, and many carry a marking. They stay inside the organisation, and each person reads only what that person is cleared to read.
What these agents do not do
No agent takes part in targeting or any use of force. None identifies, profiles, scores or predicts a person, recognises faces or infers emotion, or collects intelligence on people. None decides to arrest, charge, detain, search, admit at a border or dispatch. None handles a live emergency call, triages or dispatches: emergency services appear here for their office work only.
Agents read and draft. Officers and officials keep every decision: about a person, a record, a contract and a release.
Questions answered from the source
Staff, technicians and customs officers get the answer from your own policies, publications and guidance, with the paragraph or the page cited.
Work prepared for a decision
Maintenance records, bids, supply requests, disclosure requests and claims are checked and set out for the officer or official who decides.
Reports with the record behind each figure
Sustainment, availability and training reports are prepared from your own records, for the lead who draws the conclusions.
A record of every step
Who asked, which tool ran and what was decided.
From policy and doctrine to logistics, maintenance, procurement, records and disclosure, reports and case files, training, border and customs paperwork, people and finance.
Policy and reports
Policy questions, briefing notes, report drafts, after-action reports
Logistics and maintenance
Spares checks, supply requests, maintenance records, availability reports
Procurement and records
Tender checks, contract obligations, disclosure requests, retention reviews
Training, customs and people
Training records, inspection preparation, declaration checks, allowance claims
Each is a use case a defence, police, fire and rescue or border organisation can build in Katonic Studio. A name with a link has its own page.
Every agent follows one rule: it prepares paperwork, and a named officer or official decides.
Go deeper
Guides and answers
A defence or public safety organisation that wants AI agents on its own infrastructure has more than one kind of product to choose from. The products sold for this fall into four types: agent platforms you run yourself, infrastructure platforms, automation platforms with agent features, and developer frameworks. Our comparison describes fifteen of them with the same fields, read on each vendor’s own pages: where each runs, who operates it, governance, price and limits. For this work, running with no outside connection is the starting point: check which tool waits for a person, whose identity the agent uses, and what is recorded.
On-premises AI agent platforms compared: 15 options by type →An agent reaches your documents and systems in one of four ways, all through the same tool gateway and the same approval rules.
Connection · 01
An officer's notes, a supplier's bid, a maintenance form, a disclosure request.
Connection · 02
Policies, doctrine, standing orders, technical publications, procurement rules and training material, indexed with each reader's permissions and markings kept.
Connection · 03
↗Give Katonic the description of an interface you already run, and choose which operations become tools. This is the route to stores and spares, work orders, contracts, personnel records and the records index.
Connection · 04
↗Servers from the catalogue, deployed inside your own cluster, for the data warehouse and reporting databases. On a network with no outside connection, nothing reaches beyond it.
System names are left out on purpose. You supply the address and the sign-in. Nothing here is a tested integration with a named product.
It runs on the organisation's own infrastructure, with a model the organisation serves, on a network with no outside connection where that is required. Nothing leaves it.
Every tool call passes the organisation's policy. A policy can allow a tool, record its use, hold it for a person's approval, or block it, for the whole organisation or for one unit.
A tool that changes something waits for approval unless the organisation has decided otherwise for that tool. That is how no record, order, reply or release goes out unapproved.
No agent decides about a person, and none identifies, profiles or scores one. It prepares paperwork, and a named officer or official decides.
The agent acts as the person using it. It reads what that person is cleared to read, and no more. Where the organisation's API checks the caller, Katonic passes the signed-in user's identity.
Every call is recorded, with who asked, which tool, and what was decided, and an agent, a tool or a whole server can be switched off at once.
From visual workflows to full code - pick the path that fits your team and project.
Visual Builder
Follow guided steps to build agents. Configure each stage through the UI. No code required.
Best for
Logistics, records and procurement staff
Conversational Builder
Staff describe how they check a maintenance record against the publication or a bid against the tender's mandatory requirements. AI generates the agent. Developers refine together.
Best for
Staff + Developers together
Full IDEs
JupyterLab, VS Code, RStudio with GPU access, pre-installed frameworks, and direct deployment.
Best for
Data Scientists & ML Engineers
The rules on AI, policing, borders and personal data differ by country and point the same way. Each is a matter of what the system is allowed to do.
Some uses are prohibited outright
In the European Union some uses of AI are prohibited outright, among them predicting crime from a profile alone and scraping faces to build databases. These agents do none of them: none identifies, profiles, scores or predicts a person, and none recognises faces or infers emotion.
Law enforcement and border uses are high-risk
In the European Union, AI used in law enforcement and at borders for the listed purposes is high-risk from 2 December 2027. These agents keep to paperwork: policies, records, reports, bids and goods declarations.
A decision about a person is not left to software
A decision that significantly affects a person is not left to software alone. Here no agent decides to arrest, charge, detain, search or admit at a border. It prepares paperwork, and a named officer or official decides.
A drafted report says so
A report drafted with AI should say so, and the officer who signs it owns it. Here the draft is marked as written with an agent, the officer's notes are kept beside it, and the officer reads, corrects and signs.
Katonic AI is certified to ISO 27001. This is a summary of published rules, not legal advice.
In a support organisation's headquarters, a staff officer asks what the standing orders say about a movement request, and an agent answers from the orders with the paragraph cited; any interpretation goes to the policy owner. In its maintenance office, an agent reads a completed maintenance record and lists the missing entries, the missing signatures and the values outside the limits the publication sets, and the authorising engineer accepts or rejects the record. In a police force's information office, an agent gathers the records a disclosure request covers and lists the passages that may fall under an exemption, and the disclosure officer decides every exemption and releases the response. Each one reads what the organisation already has, changes nothing by itself, and leaves a record.
Seven questions to ask any supplier before an agent touches a record or a report, with our answers.
On infrastructure you control: your data centre, a sovereign cloud or a network with no outside connection. The model is one you serve there.
What the person using it is cleared to read, and no more. Knowledge sources keep each reader's permissions and markings, and where your API checks the caller, Katonic passes the signed-in user's identity.
Only what you allow. A tool that changes a record, a reply or a release waits for approval unless you have decided otherwise for that tool.
A named officer or official. No agent decides to arrest, charge, detain, search, admit at a border or dispatch, and none takes part in targeting or any use of force.
No. None identifies, profiles, scores or predicts a person, recognises faces, infers emotion or collects intelligence on people. The agents read documents and records for paperwork.
No. No agent takes a call, triages or dispatches. Emergency services appear here for their office work only: policy, training, records and after-action reports.
Who asked, which tool ran and what was decided, for every call. An agent, a tool or a whole server can be switched off at once by your administrators.
A policy and doctrine Q&A agent can run on the first day, because no system of record is touched. Start on unclassified or invented data, and talk to our team about building it on your infrastructure.