How can I help you today?

UT Assist is an AI assistant for IT support and can make mistakes. Do not share any privacy-sensitive data. Privacy Statement.

IT Setup
Help to connect to Eduroam

Software
Installation guides

Printing
Campus printer setup

IT Account
Password reset assistance

New Phone
Device setup

VPN Access
Connect from off-campus

Internal

Select an internal agent and send a message.

Notebook

Checklist

Quality

From To

How these numbers are measured

Self-service rate: every answer gets one verdict from the strongest signal it carries. A fallback reply or a clarifying question counts against it. An explicit thumbs down or thumbs up outranks that. Otherwise the grounding score decides, which measures how well the knowledge base supported the answer. Answers carrying no signal at all stay out of the calculation instead of counting as successes, so n tells you how many answers could actually be judged.

CSAT is the share of thumbs up among all ratings given. Well-grounded answers is the share of knowledge-base answers scoring at or above the grounding threshold. Open knowledge gaps is a current count and ignores the date range.

Treat the self-service rate as directional. It cannot see whether someone went on to file a ticket anyway.

Answer quality

—

Self-service rate — Well-grounded answers —

CSAT (thumbs up)

—

Open knowledge gaps

—

Answer outcomes per day

Satisfaction per day

Questions by language

Needs attention

How this list is built

Seven sources, a tab each: open knowledge gaps, thumbs-down answers that are not yet eval cases, agents whose most recent eval run regressed, conflicting articles, articles due for review, low-rated articles and broken links. Gaps and thumbs-down answers are tagged safe or needs you by a deterministic gate, and the bulk button actions the safe rows in one go. The other sources always need you: a regression means something that used to pass no longer does, and which article is right is a person's call.

Knowledge base health

Documents worth a look

Use counts come from the documents file search actually retrieved in the selected range. Newest and oldest are when a document was approved into its knowledge base, so they do not follow that range.

Show

Date	Knowledge Base	Score	Grounding	Question	Recommendation	Done	Pin

Date	User	Category	Rating	Feedback	Kept

How Agent Evals work

Agent Evals is a regression harness for an agent's answer quality. You curate golden cases: a representative question plus a rubric of criteria a good answer must meet. When a run is triggered (manually here, or automatically after a knowledge-base change), the agent answers each question live and an LLM judge scores it against the rubric. A case that passed the previous run but now fails is flagged as regressed. Evals only warn. They never block chat or approvals.

ok all cases passed regressed a previously-passing case now fails Score = % of criteria met across cases must_include / must_avoid / must_cite: criterion kinds the judge checks

Runs

Golden cases

Knowledge

Submit File for AI Knowledge Base

Files uploaded here will be securely staged. An administrator must review and approve them before they are ingested by the AI Agent. You can submit up to 25 files at once.

Submit for knowledge base

Provide details and context to generate a manual using the Architect-Agent.

Select Architect Agent:

Document Library

My Team

Admin Panel

Create New User

Force Password Change On First Login

User List

Group Management

Create departments, and teams inside them. A user belongs to one department and to any of its teams; that membership grants access to the Agents and Knowledge Bases assigned to those groups.

Group Name	Department	Description	Action

Triage & Fallback Tuning

Sampling for the triage router (which agent handles a ticket) and the "no answer found" reply. Leave a field blank to use the model's default; a value is sent only when one is set here.

Triage router

Temperature

Top-p

No answer found

Temperature

Top-p

Knowledge Base Retrieval

How much knowledge-base context the file-search tool pulls into each question — the main driver of input-token cost. Fewer results and a higher relevance threshold mean fewer tokens per question. Leave a field blank to use the built-in default (5 results, 0.5 threshold). Changes apply to each agent the next time it is created, edited, or re-loaded on server restart.

Max results (1–50)

Score threshold (0–1)

AI Agents

Assign a Vector Store and write custom System Prompts directly.

Agent Name	Description	Type	Model	Linked Knowledge Base	Action

System Instructions

Manage reusable system prompt templates. Updating a template will automatically push the changes to all linked Agents in Azure.

Instruction Set Name	Action

Toolboxes

Foundry toolboxes in this project. An agent can only use a toolbox once it has a remote-tool connection; set one up here.

Tools in a toolbox that sign in with the user's own account do not work in UT Assist, and one such tool breaks every answer of the agent. After changing a toolbox in Foundry, make the new version the default.

Toolbox	Default version	Connection	Action

LLM Models

Manage the list of Azure LLM Models available for your agents.

Model Name	Action

Agent Playground

Test agent instructions, tool access, and knowledge base routing in an isolated environment.

Select an agent above and send a message to start testing.

Global Security

Allow users to use the chat as a Guest without logging in. Enable Guest Access

Global Announcements

Frontpage Announcement (Leave blank to hide)

Global Routing (Default Chat Agent)

Select which AI Agent users will talk to by default on the main Chat page.

Assistant name

The name on answers, in the welcome text and in the message box. An agent can show its own name instead (Edit Agent).

Branding

The name and logo users see on every screen, in PDFs and mail, and in the privacy statement and help texts. Both host names show the same brand. The page reloads after saving.

Escalation to the service desk

Signed-in users can send a question and its answer to the service desk by email. Guests never see the button.

Show the "Contact the service desk" button Service desk email address

Mail server

Leave the server empty to use the SMTP settings from the server's environment file. The certificate is always checked; a user name and password are only sent over STARTTLS or TLS.

Server Encryption Port Sender address User name Password

Remove the stored password

Inactive account cleanup

Once a day, accounts that have not signed in for the set number of days are deleted with their data. Administrators and team managers are never deleted. In report mode the cleanup only lists who would go.

Website Crawler

Keeps the English IT pages of the websites below as Markdown in Azure Blob Storage, so an agent can answer from them. Pages that are not English or not about IT are skipped. The Azure AI Search knowledge source over that container is made outside UT Assist and attached under Search Indexes.

Crawl every day

Allowed domains Never crawl these hosts Start pages Sitemaps Skip addresses containing IT keywords

One per line.

Keywords a page needs
Page limit
Link depth
Parallel fetches
Delay between requests (seconds)
Daily start (Amsterdam time)
Include PDFs linked from IT pages

Last run

Run history

Started	Duration	Status	Fetched	Kept	Uploaded	Unchanged	Deleted

Pages in the last run

Address	Title	Status	Reason	Last modified	Action

Always kept

Address	Action

Never indexed

Address	Action

Knowledge source

Vector Stores (Knowledge Bases)

Create and manage Vector Stores in Azure AI Foundry. You can link these to specific agents in the Agent Management tab.

Store Name	Description	Access Group	Vector Store ID	Action

Search Indexes

Let an agent answer from an Azure AI Search index, such as one built over a SharePoint library. Only the pointer is stored here: nothing is copied into UT Assist, nothing is synchronised, and deleting a registration leaves the index in Azure untouched.

Name	Connection ID	Index	Query type	Top-k	Filter	Access Group	Active	Action

Recycle Bin

Deleted Knowledge Bases are kept here for 90 days before the automated Janitor permanently destroys them.

Store Name	Description	Date Deleted	Action

Questions Asked

Start Date

End Date

Questions by Language

Activity Heatmap

Questions by weekday and hour, Europe/Amsterdam local time.

Total Requests

0

Unique IPs

0

Top Endpoint

-

Traffic Overview

Operating System

Browser

Device Type

Search (IP, Endpoint, or Device)

Endpoint

Start Date

End Date

Block duration Sort by

Needs attention

Staff Activity Summary

Background work per person: admin and knowledge-base actions, following the category filter below. Chat questions and sign-ins are not counted. Audit history is kept for 90 days.

Start Date

End Date

Work per person

Work per day

By User

User	Actions	Categories

Top Actions

Action	Count

Timestamp	User	Category	Action	Details

Settings

Profile

Change Password

Two-Factor Authentication

Manage the authenticator devices enrolled on your account.

Passkeys

Appearance

Theme

Accent Color

Chat Text Size Standard

Font Family

Reset Settings

Info

Frequently Asked Questions

Usage over the last 90 days. Energy, water and carbon figures are estimates derived from the methodology shown at the bottom.

🎯 Questions answered (all-time)

—

⏱️ Time saved (all-time)

—

🏆 Next milestone

—

Questions asked (last 90 days)

—

Total tokens (last 90 days)

—

Input tokens (last 90 days)

—

Output tokens (last 90 days)

—

⚡ Energy used (last 90 days)

—

🌍 Carbon footprint (last 90 days)

—

💧 Water used (last 90 days)

—

Languages (last 90 days)

Tokens per day (last 90 days)

Questions per day (last 90 days)

Methodology & assumptions

For complex or confidential matters, you can always contact the Service Desk ICT.

UT Assist Login

Change Password

Two-Factor Authentication

Backup Code

Setup Two-Factor Authentication

Save Your Backup Codes

Reset Password

How can I help you today?

Internal

Notebook

Checklist

Quality

Answer outcomes per day

Satisfaction per day

Questions by language

Needs attention

Knowledge base health

Runs

Golden cases

Suggested Analyses

Saved reports

Analysis Results

Knowledge

Submit File for AI Knowledge Base

Review Queue

Active Database Files

Generated Manual

Save Manual as .md to Database

Upload New Documents

Document Library

My Team

Admin Panel

Create New User

Add an existing account to your team

User List

Group Management

Triage & Fallback Tuning

Knowledge Base Retrieval

AI Agents

System Instructions

Toolboxes

LLM Models

Agent Playground

Global Security

Global Announcements

Global Routing (Default Chat Agent)

Assistant name

Branding

Escalation to the service desk

Mail server

Inactive account cleanup

Automation

Website Crawler

Last run

Run history

Pages in the last run

Always kept

Never indexed

Knowledge source

Vector Stores (Knowledge Bases)

Search Indexes

Recycle Bin

Questions Asked

Questions by Language

Activity Heatmap

Total Requests

0

Unique IPs

0

Top Endpoint

-

Traffic Overview

Operating System

Browser

Device Type

Blocked in the past 90 days

Blocked addresses

Staff Activity Summary

Work per person

Work per day