AI process automation for your business.
We automate business processes, such as invoices or support requests, and systems operations on Linux, AIX and IBM i. AI comes in at the steps that have to interpret something, everything can run on your own servers, and your AI agents can later use those same automations.
Most of a workflow needs no AI.
An email arrives, the attachment is saved, a record is created in the ERP and the right person is notified. Those steps follow fixed rules, and a workflow runs them the same way every time, with no charge per model call and no answers that change from one day to the next.
AI comes in at the step where something has to be interpreted, such as pulling the amount from an invoice that every supplier formats differently, or deciding which team should get a hand-written support request. If it gets it wrong, the execution history shows at which step and with what data, and doubtful cases can go past a person before carrying on.
Examples of what we automate
In orange, the steps where an AI model does the work.
Supplier invoices
- PDF attachment
- AI reads the invoice
- Match with the order
- Posted in the ERP
The model pulls the supplier, VAT number, amount and order number from each PDF. If everything matches the order in the ERP, the invoice is posted automatically; if not, it waits in a queue for someone in accounts to check, with the reason flagged.
Support requests
- Email or form
- AI classifies it
- Ticket
- Team notified
Each message becomes a ticket with the category, priority and team already set, and the notification arrives in Teams or by email. Whoever handles support no longer has to hand out the work manually.
System alerts
- Monitoring alert
- AI suggests the cause
- Approved playbook
- Ticket closed
When monitoring flags a nearly full disk or a service that is down, Event-Driven Ansible runs the matching playbook, on Linux, AIX or IBM i, from a list reviewed in advance. AI only suggests the cause and the fix: whatever runs is written and tested, and anything not on the list waits for a person.
New starters
- Added in HR
- Directory account
- Mailbox and permissions
- Manager notified
This one has no AI. Once the person is added in the HR system, the account, the mailbox and the permission groups for their department are created, and their manager receives a list of what is already done.
The same automations, within reach of an agent
An n8n workflow can be published as an MCP tool, and Red Hat already offers, as a technology preview, an MCP server for Ansible Automation Platform. That way an AI agent asks for an existing workflow or playbook to be run instead of improvising the steps. What it can launch, with which permissions and what needs your approval is decided beforehand, and every call is logged, following the same approach we take with AI agents.
How we set it up
The right tool for the job For business processes, n8n installed on your own servers or a custom-built service, and Zapier or Make if you already use them and the data can leave the company. For systems, Ansible and Event-Driven Ansible, with AWX or with Red Hat Ansible Automation Platform. Whatever runs in-house we deploy in containers, with Docker or on your Kubernetes, and the data stays inside your network.
One account per automation Each workflow or playbook signs in to the ERP, email, directory or servers with a service account that can only do what that automation needs.
Errors you can see If an application does not respond or a field is missing, the workflow alerts whoever needs to know and the execution is kept so it can be run again.
The model, in-house too For the AI steps we serve an open model with Ollama or vLLM on your machines, which is what we do in AI inference on your own servers. An external API is also an option, as long as you know what data it receives.
Training for your team
Automations change whenever the process or the system changes, so someone in-house should know how to edit them. For systems there is the Ansible workshop for IBM Power (2 days), covering AIX, IBM i and Linux, and the Ansible and AWX course (2 to 3 days). For processes, n8n basics and advanced n8n, one day each. And if the next step is agents, the AI agents course (3 days).
Frequently asked questions
Can we keep using Zapier or Make?
Yes, if the data passing through the workflows can leave the company. They run in their own cloud and charge by the number of tasks or operations, so when customer or staff data is involved, or volumes grow, it pays to move to a tool installed on your own servers, such as n8n.
Can it connect to our ERP?
If it has an API, yes. If not, it is usually possible to read from its database or exchange files, and for writing we only use the routes the ERP itself supports. We are an IBM Business Partner, so we also connect applications running on IBM i.
Do we need Red Hat for the Ansible part?
No. Ansible and AWX are open source and cover most cases. Ansible Automation Platform is the version supported by Red Hat and the one that ships the MCP server for agents; we will tell you whether it is worth it given how many systems you have and who will maintain them.
Where do we start?
With a process that is repeated often and follows clear rules. You tell us how it is done today and which applications it touches, we build it and, once it works, we move on to the next one.
Which task is still done by hand?
Tell us about the process, which applications it touches and how many times a week it is repeated. We will tell you which part can be automated and whether any step needs AI.
