# How to add an AI chatbot to Infor LN

> Source: https://netray.co/integrations/infor-ln-ai-chatbot
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Integrations

## Short answer

Add an AI chatbot to Infor LN by pairing a chat interface with a retrieval and tool layer that reads LN data through ION API or a replica and your documents through an index. Surface it in a portal, Teams, or Infor OS, keep it read-only, and run the model on-prem if required.

## System

Infor LN

## Capability

ai-chatbot

## Intro

LN users know the pain of navigating sessions to answer simple questions: where is this order, which warehouse holds this lot, who approved this change. A chatbot lets them ask in plain language, in the tool they already use, and get an answer with a link to the LN record behind it.

A customer or supplier-facing chatbot is a separate project from an internal one. Internal users need breadth and can be trusted with detail. External users need a narrow scope, strong identity checks and answers limited to their own orders.

Below is the design for both, built around grounding: every answer comes from a tool result or a retrieved document, with the source shown, and the bot says it does not know when it does not.

## Use Cases

### Where is my order for customers

**Description:** Authenticated customers ask about their own sales orders and shipments in a portal chat, and get dates and carrier references without calling customer service.

### Supplier self-service

**Description:** Suppliers ask which purchase orders are open and what the required dates are, and confirm dates through a defined workflow rather than free text.

### Planner Q&A

**Description:** Internal planners ask about stock, supply and demand for an item and get a short answer with the numbers used.

### Shop floor how-to

**Description:** Operators ask procedure questions answered from controlled work instructions and quality documents, with revision shown.

### New starter guide to LN

**Description:** Staff ask how to perform tasks in LN sessions and get answers from your own internal documentation, not generic guesses.

### Quote and lead-time questions for sales

**Description:** Sales staff ask for current lead times and availability of standard items while on a call.

## System Surface

- ION API: authenticated REST calls to LN services for status and lookup, with credentials held by the chatbot backend only.
- Read replica or Infor Data Lake: preferred for analytical or high-volume questions so chat traffic never touches the transactional database.
- Infor OS and Ming.le style portal surfaces: where available, an embedded widget or application can host the chat window within the existing user experience. Exact embedding options depend on your Infor OS version.
- Identity: LN users and Infor OS identities, or an external identity provider for customers and suppliers, mapped to the data they may see.
- Documents: LN attachments, shared folders and document management, indexed with permission metadata for retrieval.

## How It Works

### Define audiences and scope

**Detail:** Decide who the bot is for first. Internal planners, customers and suppliers need different tools, tone and permission models.

### Build the tool layer

**Detail:** Create read-only functions for status, availability and document search, each returning concise structured data and a link to the LN record.

### Index documents

**Detail:** Chunk and embed controlled documents with metadata for revision, owner and access group, and refresh on a schedule or on change.

### Choose the model

**Detail:** Use a hosted model where policy allows, or an open-weight model on your own GPUs for private operation. Test on real questions in your language mix.

### Add guardrails

**Detail:** Restrict topics, require tool or document evidence for factual claims, refuse out-of-scope requests and hand off to a human with context.

### Embed it where people work

**Detail:** Place it in a portal page, Teams, or the Infor OS surface, and pass the authenticated user identity to the backend.

### Measure and tune

**Detail:** Review unanswered and thumbs-down conversations weekly, add missing tools or documents, and track containment of routine questions.

## Example Flows

### Customer order status

**Trigger:** Customer asks 'when will order 5521 ship?' in the portal.

**Steps:**

- Backend confirms the logged-in customer owns that order.
- It calls the order status tool and reads line dates and shipment data.
- The bot answers with the date and shows only that customer's lines.
- If the date slipped, it offers to notify the account manager.

**Result:** A routine inquiry is answered immediately without a phone call.

### Procedure question on the floor

**Trigger:** Operator asks how to rework a failed part on a given routing.

**Steps:**

- Bot retrieves the current rework instruction for that item family.
- It answers with the steps and the document revision.
- If no instruction exists it says so and routes to the quality engineer.
- The unanswered question is logged for documentation owners.

**Result:** Operators follow the controlled procedure instead of memory or a stale copy.

## On Prem

The chatbot backend, the document index and the language model can all run inside your network. Use a retrieval index such as a vector store on your servers, an open-weight model served by vLLM or Ollama, and call ION API or the replica locally. For external users, place only a thin, authenticated front end in a DMZ and keep the model and data tiers behind it. Model sizing depends on concurrency and language needs, so test with realistic peak load before choosing GPUs.

## Security

- Authenticate every user and enforce row-level access in the tool layer; the model must never be the access control.
- For external audiences, restrict answers to the user's own orders and block cross-customer queries by design.
- Keep the chatbot read-only; route requests that need changes to a human or a defined workflow.
- Log conversations with retention and privacy rules agreed with your data protection owner.
- Filter and treat uploaded files and retrieved text as untrusted to limit prompt injection.

## Options

### Vendor chat or assistant features within the Infor stack

**Fit:** Simple internal help and navigation when you are on a version that offers them.

**Effort:** Low, but limited to what is productised.

### Build in-house with a framework and your own tools

**Fit:** Teams with engineers and an appetite to maintain retrieval and evaluation.

**Effort:** Weeks to a first release, then continuing tuning.

### Partner delivery with private hosting, such as Netray

**Fit:** Sites needing on-prem models and external-facing safeguards.

**Effort:** Scoped pilot of several weeks.

## Pitfalls

- A bot that answers without evidence will invent dates; require a tool result for every fact.
- Mixing internal and external scopes in one bot risks exposing internal data to customers.
- Stale indexed documents make the bot confidently quote superseded procedures.
- No human handoff frustrates users the first time the bot is wrong.
- Measuring only deflection and not correctness hides quality problems.

## Faq

### Can the chatbot live inside Infor OS?

**Answer:** Often it can be embedded as a web application or widget, but the available extension points depend on your Infor OS and LN versions. Check with your Infor administrator; a standalone portal or Teams deployment is a reliable alternative that needs no changes inside LN.

### Can customers use it safely?

**Answer:** Yes, if identity and scope are handled in the backend. Authenticate each customer, map them to their own orders in the tool layer, keep it read-only and limit topics. Never rely on the model alone to keep one customer's data from another.

### Does it need a large GPU?

**Answer:** Not necessarily. Modest concurrency with a mid-size open-weight model can run on a single server-class GPU, while heavier loads or large models need more. Retrieval quality and tool design matter more than model size for most LN questions.

### What stops it making things up?

**Answer:** Grounding rules: the bot answers only from tool results or retrieved documents, shows its source, and says it does not know otherwise. You back this with tests on real questions and a review loop for failed conversations.

### How is this different from Power BI Q&A?

**Answer:** Power BI Q&A queries a data model for charts and aggregates. A chatbot also handles documents, procedures and single-record lookups across tools, and can hand off to people. Many sites use both for different audiences.

## Search terms

infor ln chatbot, ai chatbot for infor ln, infor ln virtual assistant, infor ln chatbot on-prem, infor ln customer portal chatbot, infor ln natural language, chat with infor ln data, infor ln conversational ai, infor ln ming.le chatbot

---
Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json