Access Request Preparation Agent · Use case
How can we use an AI agent to prepare access requests for the approver on our own infrastructure?
An access request preparation agent reads one request for access, sets it against your access policy and role catalogue, says who must approve it, and lists what is missing, with the rule behind each point. The system owner or the manager approves, and an administrator grants the access. The agent approves nothing, grants nothing and changes no account or group. It runs on your own infrastructure, so requests and directory records are not sent to a public AI service.
A request for access arrives.
Who must approve it, and what is missing?
- ✗ 01An access request often reaches the approver without the role, the reason or the end date, and the approver cannot tell from the ticket what the person holds already.
- ✗ 02Access lists and role rules show who can reach which system, and are confidential.
Where it stops for a person
Always. The owner or the manager approves, and an administrator grants.Always. The system owner or the manager approves, and an administrator grants. A request that fits the role rules is still a request: the agent says the rules explain it and the approver decides. A request for administrator rights, a request the role rules do not explain, and a request for a person who is leaving or has changed role are marked for the approver first.
It prepares the request.
The owner approves. An administrator grants.
Step · 01
A colleague or a manager raises an access request as a ticket or a form.
Step · 02
The agent reads the request: who it is for, which system, which level, for how long and why.
Step · 03
It finds the rule for that system and role in your access policy and role catalogue, and reads from the directory what the person holds today.
Step · 04
It lists what is missing, says whether the role rules explain the request, and names who must approve it under your policy.
Step · 05
The system owner or the manager approves or refuses, and an administrator grants the access in your own system.
What the system owner or the manager sees
Tap the screen to open it at full size.
Reads from · connectors
Your own APIs
The request, as a ticket or a form; your access policy and role catalogue, indexed as a knowledge source; and the directory, read through your own API with the read operations only. The catalogue has no directory connector, so you give Katonic the description of the interface you run.
What it can change
Nothing. It checks, lists and prepares. It approves no request, grants and removes no access, and changes no account, group or ticket.
Nobody has to ask.
It starts when something arrives.
Katonic calls this an event-triggered agent, one kind of workflow agent: it starts when something happens in a system, such as an email arriving, a file landing or a record changing.
What is an event-triggered agent? →The four types of agent →
01 · How it starts
When something arrives
Another system tells it that something has happened, and it starts by itself. Nobody has to watch a queue.
02 · Where your people see it
In Work Room, under Automations
Each run appears in Work Room under Automations. A run that needs a decision waits there for a person.
See Work Room →03 · Where they already work
In a screen they already use
Your developers can put it behind a screen your people already work in, so nobody has to open a new tool.
For your IT team. Another system can start a run through the REST API or a webhook. An action that is held for approval still waits for a person, however the run was started.Public API →
What you set up
- Write down, for each system, the roles, what each role may hold and who approves.
- Index the access policy and the role catalogue as a knowledge source, and import the description of your directory's API, keeping only the operations that read.
- Build the agent in Katonic Studio and test it on past requests, including ones that were refused.
- Publish it to the team that handles access requests, one system first.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service. Requests, role rules and directory records stay inside the company.
How you get it today
You build it in Katonic Studio: describe the job, connect what is named above, add test cases and release the agent once it passes them. There is no ready-made template for this job yet.
Does the agent grant the access?
No. It has no tool that changes an account or a group. An administrator grants the access after the owner or the manager approves.
Does it approve a request that fits the role rules?
No. It says the rules explain the request and names who must approve. The system owner or the manager decides.
Where are access requests processed?
On your own infrastructure, with a model you serve. Nothing is sent to a public AI service.
