ResolveGrid AI Services Administrators Guide

What AI Services is for

AI Services is where you set up and manage the content that powers your ResolveGrid AI troubleshooting agent. You add the manuals, videos, and photos for the equipment you support, tell the system what that equipment is, and then publish it so the live agent can help technicians and customers.

You can think of it as feeding and tuning the agent. The better the content you add here, the better the answers the agent gives out in the field.

Everything lives under AI Services in the left-hand menu of the Admin Portal. It has six pages, and they build on each other:

• Sources — the raw material: PDFs, Word docs, slides, videos, web pages, and 3D models.
• Assets — the physical things you support (for example, a printer). You link each asset to the sources that describe it, and add photos so the agent can recognize it by camera.
• Sites — optional groups of assets that are managed together (for example, one lab or one customer location).
• Accounts — optional groups of assets, sites, and procedures. Handy when you support more than one customer or brand.
• Publish — pushes everything you have set up out to the live agent.
• Settings — controls how the live agent behaves (its greeting, voice, which model it uses, and more).

Important:  Nothing you change in AI Services reaches the live agent until you go to the Publish page and publish a new runtime. This is the single most important thing to remember in this guide.

The AI Services menu (left) with the Sources page open.

Before you start

Signing in

1. Go to your ResolveGrid Admin Portal sign-in page (for this workspace: resolvegrid.carear.app/#/admin/login).
2. Type your work email and click Next.
3. Enter your password and click Sign in.
4. In the left menu, click AI Services to open its six pages.

If you do not see AI Services in the menu, AI is not turned on for your workspace, or your account does not have admin access. Contact your ResolveGrid administrator.

The typical setup flow

Most of the time you will work through the pages in roughly this order. The rest of this guide follows the same order.

1. Add your manuals, guides, and videos on the Sources page.
2. Create an Asset for each piece of equipment, and link the right sources to it.
3. Add a few clear photos to the asset and turn on Vision detection so the agent can recognize it by camera (optional but recommended).
4. Group assets into Sites or Accounts if you need to (optional).
5. Review the agent behavior on the Settings page.
6. Go to Publish and click Publish Runtime to make it all live.

Tip:  You do not have to finish everything in one sitting. Your work is saved as you go. Just remember to publish when you want your changes to reach the live agent.

Sources

What it is: Sources are the documents, videos, web pages, and 3D models the agent reads to answer questions and guide repairs. This is where most of your content comes from.

When to use it: Any time you have a new manual, quick-start guide, service bulletin, how-to video, or web page that the agent should know about.

Add a source

1. Open AI Services → Sources.
2. Click Add Source (top right).
3. Pick the tab that matches what you are adding: Document, Video, 3D Model, or Web URL.
4. For files, drag them onto the box or click Browse Files. For a web page, paste the address and click Find pages.
5. Leave Index for AI search turned on so the agent can answer questions from this source. (Turn it off only if the file is a workflow reference the agent should not quote directly.)
6. Click Upload. The source is added right away and then processed in the background.

The Add Source dialog, with the Document tab selected.

Processing can take a few seconds to a few minutes depending on the file. Watch the Status column, or click the small queue icon near the top to open Processing queues and see progress for indexing, workflow extraction, video, and web crawls.

What you can upload (and the limits)

Note:  Each file can be up to 1 GB. If a file is not accepted, the system tells you which formats are allowed.

Keeping sources organized

• Use New folder to group related sources (for example, one folder per product line). In this workspace, the Brother manuals live in a folder called Brother Printer.
• Select one or more rows to Move to folder in bulk.
• Use the Type and Asset filters above the list to find things quickly.

Add a website

1. Click Add Source → Web URL and paste the address.
2. Click Find pages. ResolveGrid lists the pages it can reach.
3. Review the list, keep the pages you want, and click Start Crawl.
4. You can let it finish in the background. Crawled pages show up grouped together, and you can later click Check for Updates to refresh them.

Row actions you will see

Each source row has a menu (the three dots) and status chips. Common ones include:

• View Workflow / Re-extract Workflow — see or rebuild the step-by-step procedure pulled from a document.
• Redo KB Indexing — re-read the document into the knowledge base (for example, after replacing a file).
• Extract Error Codes — pull a list of error codes out of a manual or web page.
• Process Clips / Reprocess Clips — for videos.
• Link Assets — connect the source to the equipment it describes.

Caution:  3D models (GLB files) cannot be indexed for AI search. They are used for visual reference and labels, not for answering questions.

Assets

What it is: An asset is a physical thing you support — a printer, a mower, a machine. The asset ties everything together: the documents that describe it, the photos that let the agent recognize it, and the sites or accounts it belongs to.

When to use it: Create one asset for each product or model you support. In this workspace there is one asset, the Brother HL-L8360CDW color laser printer.

The Assets page showing the Brother HL-L8360CDW, with its aliases, images, and linked sources.

Create an asset and link its sources

1. Open AI Services → Assets and click Add New.
2. Type a clear Asset Name (usually the model number) and create it.
3. Open the asset’s Settings and, under Linked sources, click Add to connect the manuals and videos that describe it.
4. (Optional) Add Aliases — other names people call it. The Brother asset uses aliases like CDW, HL-L8360CDW, and printer so the agent recognizes what a user types.

Tip:  Aliases matter a lot. If technicians call something “the big printer” or use a short code, add those as aliases so the agent knows what they mean.

Add photos and turn on Vision detection

Vision detection lets a technician point a camera at the equipment and have the agent recognize it automatically. To make that reliable, the agent needs a few good reference photos.

1. Open the asset’s Settings (the gear icon on the asset row).
2. Upload at least 3 photos taken from different angles and distances. Clear, well-lit photos work best.
3. Watch the Readiness score. When it reaches 50/100 or higher, the Vision detection → Enable switch becomes available.
4. Turn on Enable. The Brother asset in this workspace is at Readiness 70/100 with vision on.
5. (Optional) Add Labels on a photo to mark specific parts (this asset has Door and Display labels), and add Detection prompts to help the agent find named parts.

Asset settings: Vision detection at Readiness 70/100, linked sources and sites, and labelled parts (Door, Display).

Triage

Triage is a short first step the agent uses to point a user in the right direction before diving into a full procedure. The Brother asset shows “1 step + guidance” on its row. Open Edit Triage on the asset to review or change it.

Important:  New photos, labels, and links do not reach the live agent until you Publish. Uploading a photo here is not enough on its own.

Sites

What it is: A site is an optional group of assets that you manage together — for example, one lab, one building, or one customer location. This workspace has a site called brother printer lab that contains the Brother printer.

When to use it: Use sites when a technician typically works on a set of assets in one place, or when you want a shared checklist for that location. If you only support one product, you can skip sites entirely.

The Sites page with the “brother printer lab” site.

Working with sites

1. Open AI Services → Sites and click Add New.
2. Give the site a Name and create it.
3. Use Link Assets in the site’s settings to add the equipment that belongs there.
4. (Optional) Use Edit Checklist to add a checklist the agent can walk a technician through.
5. Use Import and Export to move a site (and its setup) between workspaces as a bundle.

Tip:  When a site is available, the live agent can start a session by asking the user which site they are at, then load all of that site’s assets at once. You can control this with Show Site Selector After Reset on the Settings page.

Accounts

What it is: An account is an optional, higher-level group of assets, sites, and procedures. Accounts are useful when you support more than one customer or brand and want to keep their content separate.

When to use it: Most single-product setups can leave this alone. There is always a Default account that holds everything, and it cannot be deleted.

The Accounts page. The Default account covers all assets and sites and cannot be deleted.

Working with accounts

• Click Add New to create an account, then use Account Settings to link its assets, sites, and SOPs (standard operating procedures).
• Use Agent Guidance to give the agent standing instructions or policies for that account.
• When more than one account exists, the live agent asks the user which account to use at the start of a session.

Note:  A source becomes an SOP by editing that source and marking it as an SOP, then linking it to the account. Changes to accounts and guidance take effect after the next publish.

Publish

What it is: Publish takes everything you have set up — sources, assets, sites, accounts, and settings — and pushes it out as a new “runtime” that the live agent uses. This is the step that makes your changes go live.

When to use it: Every time you want your latest changes to reach real users. Until you publish, the live agent keeps using the previous version.

The Publish page. “v1 active” is the live version; the table shows what will be published and whether each document is Ready and indexed (KB On).

Publish your changes

1. Open AI Services → Publish.
2. Review the sections — Sources & Assets, Sites, Accounts, and Publish History. Check that the documents you expect show Ready and, where needed, KB On.
3. Click Publish Runtime (top right).
4. Watch the progress bar. It moves through stages such as preparing assets, indexing the knowledge base, preparing videos, publishing, and activating.
5. When it finishes, the new version becomes the active runtime (for example, v1 active), and the live agent starts using it.

If you have unpublished changes, a banner tells you how many and lets you click to review them. Publish History lets you see past versions and, if needed, retry a failed publish.

If publishing is blocked

ResolveGrid checks a few things before it lets a runtime go live. If publishing is blocked, the message usually points to one of these:

Important:  Golden rule, one more time: if a change you made is not showing up for users, the most likely reason is that it has not been published yet.

Settings

What it is: The Settings page controls how the live agent behaves — its greeting, whether it speaks answers aloud, how it picks assets, and (in an advanced section) which AI models it uses. The subtitle on the page says it plainly: “Runtime settings for the live agent.”

The Settings page. These everyday settings are visible to any admin.

Everyday settings

These are safe to adjust and are shown to every admin:

Advanced settings (Danger Zone)

Below the everyday settings is a Danger Zone. It is hidden behind a password because the controls there change agent behavior for every user. To open it, click Unlock Danger Zone and enter the password.

Once unlocked, you can change things like:

• AI Agent Models — which model handles routing, ranking, finalizing answers, vision, and the fast agent.
• Agent Mode — how the agent thinks: Agentic Loop, Single Call, or Fast Agent.
• Retrieval Limits — how many workflows, knowledge-base results, videos, and error codes the agent pulls per answer.
• Vision scan and Confidence for Asset Vision Detection — camera matching behavior and how sure it must be.
• KB document extractor — how PDFs are converted for the knowledge base (Docling is the default).
• Demo Mode — canned photos/videos for demonstrations, and a Reset to Defaults button.

Part of the unlocked Danger Zone: AI Agent Models and the KB document extractor.

Caution:  The Danger Zone password is a light guardrail, not real security — it only keeps people from changing these by accident. Change these settings only if you understand the effect, and test with the agent before publishing.

Note:  If you need help accessing or changing settings within the Danger Zone, contact ResolveGrid Support.

Testing the agent

Every AI Services page has a floating agent button in the bottom-right corner. Click it to try the agent exactly as a technician would — ask a question, scan the equipment, and check the answers. Start a New Session to test from a clean slate.

Common troubleshooting scenarios

The agent doesn’t seem to know about a manual I added

1. On Sources, confirm the document’s Status is Ready (not still processing) and its Index shows it is indexed for AI search.
2. Confirm the source is linked to the asset (open the asset’s Settings → Linked sources).
3. Go to Publish and click Publish Runtime.
4. Test again with the floating agent.

My changes aren’t showing up for users

This is almost always because the changes have not been published. Go to Publish and click Publish Runtime. The active version chip at the top shows which runtime is currently live.

Publishing failed or is blocked

Read the message on the Publish page — it names the reason. The most common causes are a workflow that has not been published yet, or a knowledge document that is still indexing. Wait for indexing to finish (Status = Ready), or publish the workflow, then click Publish Runtime again. Use Publish History to retry.

Vision detection won’t turn on

The Enable switch stays off until the asset has at least 3 photos and a Readiness score of 50/100 or higher. Add more clear photos from different angles and let the readiness score catch up, then turn it on.

The agent doesn’t recognize what users type

Add aliases to the asset for the other names people use (model numbers, short codes, nicknames). Then publish. For camera recognition, make sure Vision detection is on and the asset has good photos.

Known limitations

• Changes never go live until you Publish — there is no auto-publish.
• 3D models cannot be indexed for AI search; they are for visual/label reference only.
• Vision detection needs at least 3 photos and a readiness score of 50/100 before it can be turned on.
• Assets support only one level of sub-assets (a sub-asset cannot have its own sub-assets).
• Uploads are capped (documents up to 20 files per batch, up to 1 GB each; detection prompts up to 20 per asset).
• The Danger Zone password is a guardrail against accidents, not a security control.
• Sites and Accounts are optional grouping tools — they don’t add content on their own.

Frequently asked questions

Do I have to publish after every little change?

You can make many changes and then publish once. But nothing you change is live until you publish, so publish whenever you want users to see your latest work.

What does “Index for AI search” do?

It builds the searchable knowledge the agent uses to answer questions from a source. Leave it on for manuals and guides. Turn it off only for a source you use purely as a workflow reference.

Do I need Sites and Accounts?

No. They are optional. Use Sites to manage groups of assets in one location, and Accounts to separate content for different customers or brands. A single-product setup can skip both.

Who can change Settings?

Any admin can view Settings, but saving changes requires tenant-admin access. The advanced (Danger Zone) settings are additionally password-protected.

How do I test before customers see it?

Use the floating agent button on any AI Services page to try the agent yourself. When you are happy, publish.