AI agent · ERP Knowledge Agent
Ask your ERP anything — in plain English
Get answers about stock, orders, customers and invoices in seconds, with the source records attached, instead of waiting for a report or the one person who knows the screens.
Quick answer
What is an ERP knowledge agent?
An ERP knowledge agent is an AI assistant that answers plain-English questions from your live ERP data, such as stock, open orders or overdue invoices, and shows the source records. It reads within each user's ERP permissions. Any change it suggests, like a credit hold, waits for a person to approve it.
What it does
Natural language ERP reporting, without new dashboards.
The agent turns a plain-English question into the right ERP queries, checks the result and explains it in one or two sentences.
Answers questions in plain English
"Which customers are more than 60 days overdue?" or "What is our stock of item 4410 across warehouses?" The agent finds the data and replies with the figures and the records behind them.
Shows its sources
Every answer links to the ERP documents, reports or tables it used, so finance and operations can check it in one click.
Respects your ERP roles
The agent acts as the person asking. Someone who cannot see payroll or margins in the ERP cannot see them through the agent either.
Drafts the next step
After an answer it can draft a reminder email, a purchase request or a credit hold, and send it for approval instead of posting it.
Works where people already are
Use it in a web app, Microsoft Teams, Slack or email, in English and other languages your team uses.
Workflow
From input to an approved update.
The agent does the work. A person approves the result. Only then does anything change in your systems.
- QuestionA user asks in plain English, in chat or email.
- AgentPlans the queries, reads the ERP, checks totals and writes the answer.
- Human approvalAny follow-up action, such as a credit hold, waits for approval.
- System updateApproved actions post to the ERP and are logged.
1. The question
Anyone with access can ask a question the way they would ask a colleague. The agent clarifies when a question is ambiguous, for example which company code or date range is meant.
2. The agent reads the ERP
It translates the question into queries against approved APIs, read-only database views or a reporting replica. It never runs open-ended queries on production tables, and every query is logged.
3. The answer, with sources
The agent cross-checks totals, explains assumptions, and returns a short answer with a table and links to the underlying records. If it cannot answer reliably, it says so.
4. Approval before any change
Reading is free within your permissions. Writing is not: any follow-up that changes data goes to the approval queue first.
Before and after
What changes for your team once the agent is live.
| Area | Before (manual) | After (with the agent) |
|---|---|---|
| Answering a routine data question | Someone exports from the ERP, builds a spreadsheet and replies hours or days later | The agent answers in plain English in seconds, with the source records linked |
| Who can get answers | Only the one or two people who know where the data lives | Anyone with ERP access, within their own permissions, from chat, Teams or phone |
| Questions waiting on IT or finance | A queue of ad-hoc requests that interrupts the experts every week | Routine questions answered by the agent; experts handle only the unusual ones |
| Hand-made reports | Recurring reports rebuilt by hand every month | Saved questions re-run on live data whenever needed |
| Follow-up actions | Emails and requests written from scratch after the analysis | Drafted by the agent and sent only after a person approves them |
We measure your own before-and-after numbers during discovery and the pilot.
Example scenario
Illustrative example · not a client result
A sales manager at a parts distributor asks, "Which of my accounts bought less this quarter than last, and why?" The agent compares quarter-to-date sales by customer, lists the twelve accounts with the largest drop, and notes that four of them have open credit holds. The manager asks it to draft check-in emails for the other eight. The drafts appear for review; nothing is sent until the manager approves them.
The same question used to need an export, a spreadsheet and half a morning.
A good fit when
- Finance, sales or operations teams ask the same data questions every week and wait for exports or reports.
- Only one or two people really know where the data lives in the ERP, and they are a bottleneck.
- Managers want answers on their phone or in Teams without learning ERP screens.
- You need answers from live data, not last month's dashboard.
Probably not the first step when
- Your ERP data is mostly incomplete or kept in spreadsheets outside the system; start with a data clean-up.
- You already have a well-used BI tool that answers the questions people actually ask.
- Nobody is willing to own the definitions of key figures such as margin or overdue.
Systems it connects to
Works with the tools you already run.
We connect at the category level, through each system's own APIs, database views or exports, so the agent fits your stack rather than the other way round.
ERP
Sales, purchasing, inventory, finance and production data through APIs, views or a replica.
Data warehouse
If you already have a warehouse or BI layer, the agent can query it for history and trends.
Email and chat
Ask questions and receive answers in Teams, Slack, email or a simple web app.
Document storage
Contracts, price lists and specs in shared drives can be searched alongside ERP data.
We are ERP-neutral. The same connection approach works with SAP S/4HANA and Business One, Oracle NetSuite, Microsoft Dynamics 365, Odoo, Tally, Sage, Epicor, Infor, and older or customized on-premise ERPs. We confirm the connection method for your version in discovery.
Security & control
You stay in charge of every change.
- Read-only by default; writes need explicit approval
- Uses the asking user's ERP permissions
- Every question, query and answer is logged
- Runs on-premise, in a private VPC or in the cloud
- Private, self-hosted models available for sensitive data
Pricing
Enterprise pilot from $7,500, or plans from $250 a month.
The ERP Knowledge Agent running on your real data, with approvals and an audit log. Two ways to start: monthly plans from $250 to $999 a month plus a one-time setup fee, or for mid-market and enterprise teams a fixed-price pilot from $7,500, then from $1,500 a month in production.
Related agents & industries
FAQ
Common questions.
Is this just a chatbot on top of our ERP?
No. A chatbot guesses from text. The ERP Knowledge Agent runs real queries against your ERP, checks the results and shows the source records, so every answer can be verified.
Can it see data a user should not see?
No. It acts with the permissions of the person asking, so your existing ERP roles still apply.
What happens when it does not know?
It says so, explains what is missing, and suggests who or which report can answer. It does not invent figures.
Which ERPs does it work with?
Any ERP we can reach through an API, database views or exports, including older on-premise systems. We confirm the method in discovery.
Can it run without sending data to the cloud?
Yes. It can run fully on-premise with a self-hosted model, which is how our own kutty.ai on-premise ERP AI assistant works.
Works with leading ERPs, CRMs and helpdesks — cloud, hybrid or on-premise.
ERPData warehouseMicrosoft TeamsSlackWeb chatCloud, hybrid or on-premiseSee an AI agent running on your ERP data in 4–6 weeks.
Fixed scope · fixed price. Human approval on every write. We reply within 1 US business day.