User Guide

Most of what you need to know during a working day is already in Aras Innovator. Finding it has meant knowing where to look, knowing how to build the query, or asking the person who does.

The Aras AI Assistant answers the question instead. You ask in plain language and it answers from your own Innovator data, with a link to the record the answer came from.

It can also change data when you ask it to, and it always asks you to approve the change before anything is written. See Changing Data on this page.

Anyone who works with Aras Innovator data. No knowledge of the data model or of any query language is assumed. If you can describe what you want in a sentence, you can use the assistant.

The assistant works from your live Innovator data and shows you where every answer came from, which makes checking straightforward. Do check, whenever the answer is going into a decision.

AI-generated output must be reviewed and validated by a qualified person before it is used in any operational, engineering, manufacturing, compliance or business decision. Understanding an Answer describes what to look at, and Terms of Use gives the requirement in full.

The assistant becomes available once your organization’s tenant is configured and you have been granted access. If you cannot see it at all, or you can sign in but no agents are listed, contact your Edge AI Administrator.

You can use the assistant in two places: inside Aras Innovator, alongside whatever you are working on, or on its own when you are not in Innovator at all.

Go to edge.aras.cloud/ai. The home page opens with a greeting that addresses you by name and shows the name of your tenant. A Start a conversation section lists the agents published for you. Choose one and the conversation opens.

Most organizations publish the Aras AI Assistant, which answers general questions about Aras Innovator data. Some publish additional agents for particular kinds of work. Where you see more than one, the description under each name tells you what it is for.

If you are an Edge AI Administrator, the home page also shows a More for you section with a Manage agents card. Selecting Go to Agent Library opens the Agent Library. The Administrator Guide covers configuration.

Open the assistant panel and it appears beside your work, resizable and dismissible at any time.

Everything you can do in the standalone chat you can do here, and one thing more: the panel knows which record you have open and uses it as context, so you can ask about the item in front of you without naming it. See Working in Context on this page.

For a longer session, expand the panel to full screen. Your conversation and your context come with you, and collapsing it returns you to the panel with everything intact.

• Use the embedded panel when your question is about the record in front of you, or when you want to move between answers and records without leaving Innovator.
• Use the standalone chat when you are not working in Innovator, or when your question is not about anything you have open.

The one real difference is item context. Only the embedded panel can see what you are looking at.

Where more than one agent is published, the agent name in the header lets you change to another one.

Switching starts a fresh conversation with the agent you choose, so if your current conversation already has messages you are asked to confirm before it is cleared.

Open the assistant. With nothing in the conversation yet you see its name, a short description of what it does, and a few suggested questions your administrator has set up. Those suggestions are worth reading, because they tell you what this particular agent is good at.

Select one, or type your own. Something similar to:

Show all released Parts changed in the last 30 days

An answer comes back as a list, and each row links to the record in Innovator. That is the whole loop: ask, read, follow the link to the source.

Ask in the words you would use with a colleague. You do not need to know which item type holds the answer, which properties are involved, or how the data is related. The assistant works that out and tells you what it did.

Finding and filtering items. Retrieve items by state, date, owner, attribute, or any combination.

• List all open ECOs assigned to me
• Show released Documents changed in the last 30 days
• Find Parts with no preferred manufacturer

Product structure. Explore assemblies and trace where a component is used.

• Show the BOM for ASM-7700 at rev C
• Where is Part P-4420 used?
• Which assemblies contain capacitor C-100?

Change activity. Review what a change affects, who raised it, and what is in flight.

• What is affected by ECO-2024-0451?
• Who submitted CR-2025-0033?
• Show change requests opened this quarter

Counts and totals. Get a number, or a breakdown.

• How many Parts are in this assembly?
• How many open changes do we have, by type?
• Count Documents per classification

Status and ownership. Check what is waiting on you or on someone else.

• What is assigned to me right now?
• Show items waiting on my approval
• List Parts created by Riley Miller

History. Retrieve what happened to an item, or how it has been revised.

• Show the full history of Part MP0101
• How many revisions does Part MP2954 have, and what state is each in?

Use full history or action history when you want the record of what was done.

A shorter phrasing such as “history of Part MP0101" may return revision information instead. If that happens, ask again for the action history.

The assistant can draw what it found as well as list it.

• Diagram the relationships for ASM-7700
• Show me the data model around Part and Document
• Show a pie chart of Parts by classification

Ask for the form you want in your question, or ask afterwards about a result you already have on screen.

A prompt that works usually carries three things:

• the object you are asking about
• the action you want performed on it
• the scope or format of the answer you want back

Show released Documents changed in the last 30 days

That is action, object, scope. Drop the scope and you get every document. Drop the object and the assistant has to guess what you meant.

You do not need all three every time, and a follow-up question often needs none of them because the conversation already supplies the context. But when an answer is not what you expected, the missing piece is usually one of the three.

• Name the item type when your question could apply to several.
  
  • Show all open ECOs is clearer than show all open changes because the second could mean several things.
• Narrow with what you already know. A date range, a state, an owner or a classification will get you a shorter and more useful answer than a broad question and a long list.
• Refine rather than restate. Once you have an answer, build on it with a follow-up instead of writing a new question. See Working in Context on this page.
• Watch what the assistant says it resolved. If you use a term the system does not recognise, the assistant tells you what it matched, what it could not confirm, and answers with what it was able to establish. That message is the fastest route to a better question, because it tells you exactly which word to change.

The shape of your question decides the shape of the answer. These examples cover the common ones.

Ask in whichever language you work in. The assistant understands the languages the underlying model supports and answers in the language you used. You can also ask it to reply in a different language at any point in the conversation.

Your PLM data comes back as it is stored. Item names, descriptions and property values are not translated; only the assistant’s own explanations are.

An answer is not only the result. It arrives with the record links behind it and with the assistant’s own working, and both are there so that you can check it.

Responses come back as formatted text: paragraphs, lists and tables, with a diagram or a pie chart where you asked for one.

Tables are part of the response rather than a separate grid, so they are for reading rather than for sorting or filtering. If you want the same data ordered differently, or narrowed, ask for it. That is usually quicker than rearranging it yourself would have been.

Every answer carries links to the Innovator records behind it.

In the embedded panel, following a link opens the record in Innovator and your conversation stays exactly where it was. In the standalone chat, the record opens in a new tab.

Following one link is the fastest check you can make on any answer.

Every response carries its reasoning with it, collapsed beneath the answer. Expand it and you see the whole working: the query that was run, what each tool was asked for and what it returned, and the assistant’s own description of what it was doing at each stage.

This is not diagnostic detail meant for support. It is the answer’s evidence. Open it whenever a result surprises you, because it almost always shows one of two things: that your question was read differently than you meant it, or that the data really does say what the answer says.

Where the agent plans before it acts, the plan appears in a side panel: the steps it intends to take, and the state of each one as it works through them. You can watch it progress and see exactly where it has reached.

Retrieval is bounded, so a very broad question can come back with part of the answer rather than all of it. When that happens the assistant says so in the answer, and asks you to narrow the request.

Take it at its word. A narrower question does not just avoid the limit, it usually gives you a more useful answer.

Three things, and they take about a minute between them.

• Open the reasoning and read the filters. Confirm the assistant narrowed the data the way you intended. This catches most misreadings.
• Follow one link. Open a single record from the result and confirm it says what the answer says.
• Read the count. Where the assistant has calculated across a set, it tells you how many records it used, because records missing the data it needed are left out of the calculation. If that number is lower than you expected, that is the reason, and the answer is describing a smaller set than you had in mind.

The assistant keeps track of two different things: the record you have open, and the conversation you are having. They behave differently and it is worth knowing which is which.

In the embedded panel, the assistant sees the record you are looking at in Innovator. A chip above the input shows which one, so you always know what “this” refers to.

That means you can ask about the item in front of you without naming it.
With ECO-2025-0100 open in Innovator
You: Who submitted this?

Open a different record and the chip follows: your next question uses the new one.
If your question is not about the record on screen, remove the chip and ask without it.

Follow-up questions build on the answer before them, so you do not have to restate anything.
You: Show all open ECOs assigned to me
Twelve ECOs come back.

You: Which of these were created in the last 7 days?
The list narrows to three.

You: Show the affected items for the first one

Each question does less work than the one before it, and gets you further. This is usually faster than writing one long question, and it lets you change direction when an answer is not what you expected.

If a question could mean several things, the assistant asks before it goes looking. Ask for the latest change without saying what kind, and it may ask whether you mean an ECO, an ECR or an ECN.
Answer it and the assistant continues with the narrower question.

If an answer is going somewhere you do not need, stop it with the Stop button in the input area. The assistant halts, and you can ask something else straight away.

New Chat clears the current session: the conversation and everything the assistant was holding from it.

It does not change what you have open in Innovator. If you are working in the embedded panel with a record on screen, that record is still your context, because the panel reads it from Innovator rather than from the conversation.

The assistant can change Innovator data, not only read it. Nothing is written without your explicit approval, every time.

With the right permissions and configuration, the assistant can create a record, update properties on an existing one, move an item to a different lifecycle state, or delete an item.
If your assistant cannot do these things, your organization has not enabled them. Ask your Edge AI Administrator.

When the assistant is ready to change something, it stops and shows you an approval card before anything is written.
The card tells you what it intends to do, which item it will act on, and what will be written. For an update it shows the current value alongside the proposed one, so you can see exactly what changes.

Approve and the change is submitted to Innovator under your own identity.

• If Innovator accepts it, the change is applied and the assistant confirms it, with the updated record.
• If Innovator refuses it, because your permissions do not allow that operation or because a validation rule rejects it, nothing is written and the assistant tells you why.

Reject and nothing is written. The action ends there and the conversation carries on, so you can correct what you asked for and let the assistant propose something different.

Approving authorizes the attempt, not the result.
The change runs under your own Innovator identity, so it can do exactly what you could do by hand and no more. Your permissions, the lifecycle rules and any workflow constraints all still apply.
A change made this way appears in Innovator as your change, in the record’s own history.

There is no undo. A change committed through the assistant is a change in Innovator like any other.
You can ask for the reverse as a new change, giving the previous value, and it goes through the same approval. The previous value is in the record’s history if you no longer have it.

A few things behave in ways worth knowing before you meet them.

• Asking for history may give you revisions. A short phrasing such as “history of Part MP0101" can return revision and generation information rather than the record of what was done. Ask for the full history or the action history, or just say so in your next message.
• Some items show an identifier rather than a name in the context chip. The chip may show an item type and a number instead of a readable name. The context is still correct; only the label is affected.
• Stopping a response ends that turn completely. If you stop the assistant mid-answer, the next question in the same conversation starts clean. If the assistant then behaves differently than it did before you stopped it, start a new conversation.

Your Edge AI Administrator can resolve most things faster than support can, because they are configuration rather than faults.
Contact them when the assistant is not there at all, when you sign in and no agents are listed, or when the assistant cannot answer about data you know exists. That last one usually means the data has not been made available to the assistant yet, which is something they can change.

Documentation and the Aras subscriber portal are both reachable from the help menu.
You can raise a problem in the subscriber portal yourself. Include the agent you were using, the question you asked, what came back, and what you expected instead.

AI-generated output requires human review. AI-generated outputs must be reviewed and validated by a qualified person before use in any operational, engineering, manufacturing, compliance, or business decision. See the Aras InnovatorEdge Product Specific Terms for full details.

Customer data is not used for model training. Aras does not use customer data, user prompts, inputs, or AI interaction data submitted through InnovatorEdge services to train, retrain, or improve any general-purpose or shared AI or machine learning models. See the Aras InnovatorEdge Product Specific Terms for full details.

Session data is not retained. The assistant maintains short-term context within an active conversation so that follow-up questions work. That context is not stored or retained after the session ends.