Loading
Loading
AI Agent
The AI Agent reads only what the signed-in user is allowed to see in SAP Business One. It summarises what matters, points at what looks wrong, and drafts the work — while the decision to commit anything stays with a person.
The AI Agent is part of Z-SYSTEM, not a separate system. Your use of SAP Business One and any integration remains subject to your organisation’s licence agreement.
Capabilities
Every capability runs on the same records the user already sees on screen. There is no shadow copy of your database and no retrieval across permission boundaries.
Type a question in plain language and the agent retrieves the answer from the documents and master data your account can access. Each answer lists the records behind it, so you can verify the source before you act on it.
“Which customers have the largest overdue balances this month?”
Open your screen in the morning and see what is waiting for approval, what is close to its due date, and what carried over from yesterday. The briefing follows the reader’s role, so a warehouse manager and an accounting manager do not get the same list.
“Give me the purchasing team’s briefing for this morning.”
The agent compares new entries against the usual pattern for that customer, vendor or item and flags what stands out — an unusual price, an unusual quantity, a possible duplicate. Flags are raised for a person to review, never acted on automatically.
“Which purchase orders this week were priced outside our usual range?”
Draft a document from an existing one, fill in the fields the system already knows, and check what is still missing before it moves on. Everything stays a draft until a user confirms it.
“Draft a quotation from sales order SO-2417 using the same payment terms.”
Answer questions about stock by warehouse, items below their reorder point, and quantities that have not moved for months. Each figure links back to the receipts and issues behind it, so you see the reason and not just the number.
“Which spare parts are below their reorder point in warehouse WH-01?”
Pull a customer’s buying history, pricing terms and open document status into one view before a call or a meeting. Orders that are late or approaching their delivery date are called out first.
“Summarise the Northwind Trading Co. account before my meeting this afternoon.”
Compare the prices and lead times you have actually received from each vendor and read them as a simple table before raising an order. The agent also flags purchase orders that are past their receipt date with nothing booked in.
“Which purchase orders are past their receipt date with no goods receipt?”
See exactly where a request is sitting, who it is waiting on, and how long it has been there. The information an approver needs to decide is gathered onto a single screen.
“Where is the production team’s purchase request sitting right now?”
When SAP Business One rejects an entry, the agent translates the message into language a business user can act on and names the field or document that caused it. It then suggests a fix the user can apply themselves.
“Why did delivery note DN-0771 fail to post?”
Open work is ordered by its effect on delivery, cash collection and the nearest deadline, rather than simply by arrival time. The order is a suggestion, and the user can always override it.
“If I have one hour to clear work, what should I start with?”
Anything that changes data in SAP Business One stops at a confirmation screen first. The user sees the full set of values about to be written, then confirms, edits or cancels.
“Review the drafted purchase order, then confirm it yourself to post it.”
The agent runs with the same permissions the user already has on screen, with no hidden elevated access. If a user cannot open a record themselves, it will not appear in the agent’s answer either.
“A sales rep asks for item cost and is told it sits outside their access.”
Every question, draft and confirmation is recorded with the user, the timestamp and the records used. An administrator can reconstruct where any given answer came from.
“Show me who confirmed the drafts the agent prepared yesterday.”
Try it
Pick a scenario below to see the shape of an answer: the reasoning in plain language, the evidence attached to it, and the actions a person still has to confirm.
Z-SYSTEM AI Agent
This demo runs entirely on scripted replies and invented sample data. No language model is called and nothing on this page is connected to SAP Business One. In a real deployment you connect the AI provider your organisation has chosen.
Governance
An AI assistant is only useful if the organisation can trust what it is doing. The rules below are enforced by the system, not offered as usage guidance.
Every question is answered under the signed-in user’s own permissions. Records outside that scope are never retrieved in the first place — they are not fetched and then hidden.
The agent can only produce drafts. Posting a document, approving a request or changing data in SAP Business One always requires an explicit confirmation from a user.
Every step with a binding effect pauses for a decision by someone authorised to make it. There is no mode in which the agent keeps working unattended.
The question asked, the draft produced, who confirmed it and when are all written to an audit log that administrators can review and export.
Only the records relevant to the question are sent to the AI provider, never a copy of the database. Your organisation decides which fields may never leave the system at all.
Each answer carries the documents and figures it was built from. Users can open the source directly instead of having to take the answer on trust.
AI providers
Z-SYSTEM talks to models through one pluggable interface, so changing provider is a configuration decision rather than a rebuild.
Works with any provider that speaks the widely adopted OpenAI-style API. The endpoint and model name are set server-side, so you can point at a different service without touching the application.
A natural fit for organisations already standardised on Microsoft Azure that need processing to stay in a named region. Both the target resource and the deployment name are configurable.
For organisations whose data must not leave their own network, Z-SYSTEM connects to an LLM running on your infrastructure through the same interface. No content is sent to a third-party service in this mode.
With no provider key configured, the agent runs in mock mode on scripted responses. Teams can review the screens, the wording and the approval flow before committing to a model.
Permission checks, human confirmation and audit logging live inside Z-SYSTEM rather than in the model provider. Swapping models therefore leaves your control rules exactly as they were.
Provider API keys are held and used server-side only. The browser never receives a key and never calls a model provider directly. The service is built to security best practice, with additional controls configurable to your organisation’s policy. Results depend on your processes, existing systems and the scope of each deployment.