Kad¶

Kad is Kadratis's built-in chat assistant. It is connected to the same data you already work with (applications, assets, technologies, incidents, projects, requests, tasks, contracts, knowledge documents, and master data), so you can ask questions in plain language instead of clicking through several screens to find an answer.

Kad does not replace the rest of the application. It is a faster way to summarise context, locate records, draft text, or prepare changes. Sensitive operations are always shown as previews you must approve before anything is written back into Kadratis.

Kad vs AI Agents

Kad is the interactive assistant you drive: you ask, it answers, and you approve one change at a time. AI Agents are different: they are autonomous helpers that watch your service desk and propose or handle ticket work on their own, within the approval limits you set. See AI Agents: Overview.

Where to find it¶

Workspace: Kad (top navigation)
Route: /ai
Permission: ai_chat:reader lets you open the chat workspace and start conversations
Feature flag: requires the chat surface to be enabled on your instance. If it is turned off, the workspace shows a notice instead of the chat UI.

The workspace is also available to administrators with ai_chat:admin, who can see and moderate everything regular users do.

Before the included model is accepted¶

On the hosted service, Kad can run on the Kadratis included model. Before it does, an administrator accepts the provider and where it processes data. New trial workspaces start in this state, with Kad turned off.

Until then, Kad's page opens normally with its suggestions. A box takes the place of the message field: I agree that {provider} ({location}) processes my questions and the data they concern for Kad's answers. They are not used to train its models., followed by Another model, or turn it off: Admin › Kad.

An administrator of the Kad settings ticks the box. This turns Kad on and records the acceptance, with who accepted and when. The message field then appears.
Anyone else reads Kad is waiting for an administrator to turn it on.

Administrators can also accept on the Kad Settings page. Organizations that use a model of their own never see this box.

Chat turned off

When an administrator turns chat off under Administration › Kad, a person who may use the chat still opens this page, and the Kad entry stays in the top navigation. The page says Kad is turned off for this workspace. An administrator of the assistant also sees Turn it on in Administration › Kad, which opens the Kad settings. Anyone else sees Ask an administrator to turn it on. A person without the chat right sees the access denied page. While the included model waits for its acceptance, the page shows the acceptance box instead, even with chat turned off: see Before the included model is accepted.

Starting a conversation¶

When you open Kad for the first time, you land on a welcome screen with:

A short tagline ("Kad is ready") and a description of what you can ask
A Try asking section with sample prompts you can click to send
The composer at the bottom, ready to accept your first message

Click any suggestion to either send it directly, or (for prompts that contain an @ placeholder) drop the suggestion into the composer so you can finish it.

When a path was chosen for the workspace at its first opening, during the trial, the suggestions start with its two questions, among those you may read, then the usual ones, five at most:

Path	Questions
IT budget	What are the main savings opportunities in IT OPEX?, Which budget lines have no owner?
IT landscape map	Which technologies reach their end of support within six months?, Which critical applications have no recovery plan?
Projects and requests	Which projects are over budget?, Which requests are waiting for a decision?

Type a message and press Enter to send. Shift+Enter inserts a new line. The send button is disabled while there is nothing to send and morphs into a red Stop button while Kad is responding.

The conversation list¶

The left sidebar lists your past conversations. Use the menu icon in the top-left of the chat area to collapse or expand it.

The list contains:

A New conversation button at the top
A search field that appears as soon as you have at least one conversation
Conversations grouped by date: Today, Yesterday, Last 7 days, Older

Each row shows the conversation title (or Untitled when none has been set yet). Hover or focus a row to reveal:

A pencil icon: Rename the conversation. Double-click the title to do the same thing.
A trash icon: Delete the conversation. The conversation and its messages are deleted. This cannot be undone. If you delete the conversation you currently have open, Kad switches to a fresh empty chat.

Search filters the list by title as you type. Renaming and deleting are saved immediately.

Writing a message¶

The composer is the main control point of the workspace. It supports:

Multi-line text up to 10 visible rows before it scrolls
Attachments: images (PNG, JPG, GIF, WEBP) and spreadsheets (.xlsx, CSV)
@-mentions of Kadratis records
A keyboard hint reminder ("Enter to send · Shift+Enter for new line")

Attaching images and spreadsheets¶

You can attach images (PNG, JPG, GIF, WEBP) and spreadsheets: Excel workbooks (.xlsx) and CSV files. There are three ways to add them:

Click the paperclip icon (Attach a file) and pick files from your computer
Drag and drop files onto the composer (a hint overlay confirms the drop target)
Paste a file directly from the clipboard

Each pending image appears as a thumbnail above the text. Each pending spreadsheet appears as a small tile with its name and size. Click the small X to remove one. A message holds up to 8 attachments, images and spreadsheets together. Once the limit is reached, the paperclip is disabled until you remove or send the current attachments.

Images are uploaded with your message so Kad can describe them, compare them, or extract details.

Kadratis reads a spreadsheet when you send the message:

A file can hold up to 5 MB and 20,000 rows. Only the first sheet of a workbook is read.
A file that cannot be read (damaged, empty, old .xls format, too large) is refused at once with the reason. The message is not sent.
Kad receives a summary of the file: the column headers, the type of each column (text, number or date), the first ten rows and the number of rows. It can describe the file, say what each column holds and spot problems in the first rows. It cannot answer about the rows after the first ten.
The summary stays attached to the message for the rest of the conversation.

Importing a spreadsheet¶

Attach a spreadsheet and ask Kad to import it into a Kadratis list: applications, assets, projects, requests, tasks, incidents, budget lines, contracts, suppliers, contacts, companies, departments, cost centers, accounts, analytics dimension values, business processes, working-day calendars, or the technology links of servers and applications. Kad matches the columns of your file to the columns of the list's import and shows a preview:

Each column of the import, with the file column it reads (or one value for every row), how sure Kad is and how many values it read. Then the values it could not read, the file columns it leaves out, what it supposed, and the first rows as they will be imported.
The Import check: the same check as the list's own import. It counts the rows to create and to update, and lists the errors and warnings with the line numbers of your file.
When the check finds errors, Approve is unavailable. Tell Kad what a value stands for, or correct your file and attach it again. Kad prepares a new preview, which replaces the previous one.
Approve runs the import with your own rights, as the list's import would. You need the same right as for that import: applications:admin to import applications, incidents:contributor to import incidents. Read-only access is refused. The preview then shows what was created and updated.
For applications, assets, projects, requests, tasks and incidents, a blank cell keeps the value already stored on a record the file updates.
Contracts, suppliers, contacts, companies, departments, cost centers, accounts, analytics dimension values, business processes, working-day calendars and technology links replace every value of a record they update. When rows of the file match existing records and the file lacks some columns of the import, the check reports an error and the preview cannot be approved, because those values would be cleared. Map the missing columns, or keep only new records in the file.
The Download file (.xlsx) button, next to Reject, gives the file Kad prepared, as an .xlsx file in the format of the list's import: the columns of the import, your values as the import reads them (list codes, dates), one row per row of your file. For the imports that replace every value, the file holds every column of the import, blank where your file had none. Fill in the missing values and import the file with the list's own Import button. The Audit log records each download, as for any export.
For budget lines, every row creates a new line. The suppliers the file names must already exist: Kad does not create them.
The Audit log records each import made through Kad, with the file name and the counts, under the source AI chat.
The AI model receives only the summary of the file (headers, column types, first ten rows) and the counts of the check, with the code and the column of each error. The other cells and the messages of the import stay on the preview.

Mentioning records with `@`¶

Typing @ opens the mention picker above the composer. It lets you reference any Kadratis record you have access to, with two complementary modes:

Type-token prefix: short codes that map to a single entity family. Examples:
@T-5: task with the reference T-5
@DOC: recent knowledge documents
@APP backup: applications matching "backup"
@BL-12: the budget line with the reference BL-12. @BL alone lists budget lines of both types. The former numbers still work: @OPX-12 or @CPX-3 finds its line
@PRJ, @REQ, @INC, @TEC, @AST, @CONN, @INT, @LOC, @CTR, @COMP, @CONT, @DEPT, @SUP, @BP
Plain text: anything else (@payroll, @server-2) runs a cross-type search ranked by relevance. It also finds budget lines by the names of their analytics values.

Use the arrow keys to move through the suggestions, Enter or Tab to confirm, Escape to dismiss the picker. Results are grouped by entity type (Knowledge, Tasks, Projects, Applications, Assets, Contracts, and so on) so you can tell at a glance what kind of record you are about to insert.

When you confirm a suggestion, the composer keeps showing a readable label (@DOC-152, @SAP S/4HANA). When the message is sent, each mention is expanded into a real link Kad can follow back to the source record.

Editing or regenerating a message¶

Hover any of your previous messages to get message-level actions:

Copy: copy the message text to the clipboard
Edit: re-open the message in an inline editor; saving sends the new version and truncates everything that came after it (the conversation reruns from that point)
Regenerate (on assistant replies): ask Kad to produce another answer to the same prompt

Editing is the right tool when you realise your earlier question was unclear. Regenerating is the right tool when the question was fine but the answer was not.

How Kad replies¶

Kad streams its answer character by character. While streaming:

The composer stays usable so you can prepare a follow-up
The send button shows a red Stop icon. Clicking it cancels the in-flight response
A small "Using tools…" indicator appears when Kad is searching Kadratis, fetching a document, or running another tool call
The number and type of tools used are summarised under the answer once it is complete

When the stream finishes, focus jumps back to the composer so you can keep the conversation moving without reaching for the mouse.

Images stored in Kadratis, such as attachments, display as usual. An image hosted on another website does not load by itself, in an answer, in your own messages or in the preview of a proposed change. It shows as a link with an image icon: click it to open the image in a new tab. Nothing is requested from that website until you click.

Tool calls¶

Kad uses a small set of internal tools to answer questions: Search all, Search knowledge, Get document, Get entity context, and a handful of others. Each tool call shows up as a compact line under the message ("used Search all · 8 results"). You usually do not need to read the tool details, but they are there if you want to see exactly which records the answer was based on.

When you ask how to do something in Kadratis, such as where a setting is or the steps of a task, Kad reads the user manual of your Kadratis version, in your language (Browse the user manual, Read the user manual). It describes the screens as the manual does and links the section it read on doc.kadratis.com.

Search all finds budget lines by the names of their analytics values, and tells Kad which values each line holds. Kad names a budget line by its BL-12 reference, and its links open the line. A number alone, such as 21, means BL-21. The former numbers still work: OPX-12 or CPX-3 finds its line.

Kad reads OPEX and CAPEX lines together. A question that names neither type covers both, and Kad can filter or group the lines by type.

Kad lists and reads technologies for users who can read the infrastructure. It can filter them by family, publisher, version, owner, end of support or number of servers, and read a technology's versions, servers, applications, owners, support contacts, web links and vulnerability alert settings. It also reads the two views of the End of support report and the version and dates of each application. Asked about a server, an application or a technology by its name, without its reference, it also reads the related records. Ask, for example: "Which technologies have no IT owner?" Assistants connected over MCP get the same answers.

When vulnerabilities are followed, Kad also lists, counts and reads the vulnerabilities, for users who can read the infrastructure. One item is a CVE on one version of a technology, as on the Vulnerabilities page. Kad can filter them by severity, exploitation, state, alerts, technology, publication date, asset or application. It can also keep the technologies in use in the estate, or the rows that no longer match. It counts them by severity, technology or version, in rows and in distinct CVE, as the page does: "117 rows (112 CVE)". Unless you ask otherwise, Kad keeps the vulnerabilities still to review that are in the alerts, as the page does. It also says when Kadratis last read the NVD and CVE.org, and whether the last read failed. The servers and applications a vulnerability reaches follow direct links, as the page counts them (see What Kadratis counts). On assets and applications, Kad reads how many vulnerabilities are still to review and how many are critical or high. Ask, for example: "Which critical applications are reached by an exploited CVE?" Assistants connected over MCP get the same answers.

Kad lists, counts and reads incidents. It filters them by asset, application or technology, groups them by linked record or by month, and reads the time to resolution. A restricted incident you cannot see appears nowhere. See Ask the assistant.

Kad reads contracts as the Contracts list shows them: the state of the current term (to decide, in progress, renewed, ends without renewal, expired), the next due date and the cancellation deadline, with the decision window of your workspace and the renewal decisions already recorded. It also reads what a contract covers: its applications, assets, technologies and budget lines. Ask, for example: "Which contracts are to decide, and which applications do they cover?" or "Which contracts cover APP-8?" Assistants connected over MCP get the same answers.

Kad names a record by its reference, such as APP-8 or BL-14, or by its name when it has none, such as a contract, a supplier or a site. The reference or the name links to the record's page when you can open it.

Kad names a record linked to the one you ask about only when you can read its module, as the record's page does: the applications of a project need the applications, its assets the infrastructure, its source requests the requests. A Business Contributor only gets the projects, requests and applications they take part in. The servers an application runs on are the exception: they come with the application, as on its Deployments tab. Asked about a person, Kad gives what the users list shows you: an administrator of users gets the full profile, everyone else the name, email and status. Assistants connected over MCP get the same answers.

An administrator of users with access to all data reads with Kad what Admin › Roles and Admin › Data access show. Kad lists the roles, the number of people in each and the rights each role gives, with the words of the Roles screen. It says who is limited to which companies, cost centers or groups, in Read or Edit, who is limited with no node, and who sees a company. The users list also carries every role of a person and their access. Ask, for example: "Who is limited to Alpha Spain, and in which access?" or "Which roles give the Admin level on budget?" A person whose own access is limited does not read these, even as an administrator of users, and Kad says so. Kad changes none of them: a person's data access, their roles, the rights of a role, and creating or deactivating a person stay on the Admin screens. Assistants connected over MCP get the same answers.

Kad reads the history of a record, as its History tab shows it, for anyone who can open the record. Ask, for example: "Who changed the amounts of BL-12?" A record without a History tab, such as a user or a role, has no history in Kadratis: Kad says so and does not rebuild one from other data. Assistants connected over MCP get the same answers.

When Kad adds up budget amounts, a line whose currency has no exchange rate for the year is left out of the sum. Kad then says which currencies and years it left out. Assistants connected over MCP get the same answer.

Kad reads the staff figures of the Cost per FTE report for each year and budget column: the cost and FTE of the lines in people or days, the cost of the lines priced per day and the days they buy. Kadratis computes the cost per FTE and the average daily rate per supplier, company, cost center or any other grouping, as the report does, and Kad quotes them. Ask, for example: "What are the external FTE and the average daily rate per supplier in 2026?" When you name neither OPEX nor CAPEX, Kad answers for both types. Assistants connected over MCP get the same answers.

Kad also gives the figures of other budget reports. With read access to accounts, it totals the budget by consolidation account, as the Consolidation accounts report does. Without it, Kad says that this needs the right on accounts. It measures several columns or years in one answer and adds a second grouping, for example the budget, revision and expected landing of each company, or each supplier over three years. It finds the largest increases or decreases between two columns, 10 by default, as the Top increase / decrease report does. It gives the FTE month by month with the average and the peak month, as Staffing by month does. Ask, for example: "Which lines rise the most from the 2026 budget to the revision?" or "What is the budget of each supplier in 2025, 2026 and 2027?" Kad reads every budget line, while the reports read the lines still active on 1 January of the year before the first year shown, so a line that ended earlier can make a figure differ from the report. Assistants connected over MCP get the same answers.

With read access to Reporting, Kad reads the chargeback as the Global chargeback and Company chargeback reports show it: the amount charged to each beneficiary company and department, what each company pays, the net, the ratios per user and to turnover, and the lines charged to one company. When you name neither OPEX nor CAPEX, it covers both types. Kad repeats the warnings of the report, such as a line it could not spread, a line without a paying company or an amount without an exchange rate. When your data access is limited, the figures cover the lines you can read, without the group total, and Kad lists the lines charged to a company only for the companies of your access. Assistants connected over MCP get the same answers.

Kad reads four kinds of reference data, as their screens show them. It changes none of them.

Cost centers, with read access to cost centers: the tree of groups and cost centers, the company and the budget holder of each node, and the number of OPEX and CAPEX lines that use it. Kad lists every node under a group, at every level, or only the nodes directly under it.
Working-day calendars, with read access to working-day calendars: the working days of a year, month by month, and the public holidays of a standard calendar. Kadratis computes the total of the year, and Kad quotes it. A country without a calendar in your workspace has none, and Kad does not compute its days.
Analytics dimensions, with read access to analytics: the settings of each dimension (the line types it applies to, whether a value is required, whether anyone can add a value), its number of values and the lines that use it. A required dimension counts as required only while it is enabled.
Exchange rates and the reporting currency, with read access to the budget: the reporting currency, the default and allowed currencies, the rate of each currency for each year with its source, the rates entered by hand and the missing rates.

A rate is the amount in the reporting currency for 1 unit of the currency. In a workspace that reports in euros, a rate of 0.89 for USD means 1 USD = 0.89 EUR. Kadratis computes the inverse rate and the rate between two other currencies, such as USD to CHF, and Kad quotes them.

Without the right a kind of reference data needs, Kad names the missing right. When your data access is limited, you read all the reference data, as on the screens. Only the counts of budget lines cover your access, and the currencies listed with their rates are those of the lines and contracts you can read.

Kad reads the history of a cost center, a calendar or a dimension as its History tab shows it. Search all finds a cost center, a calendar or a dimension by its code or its name. It does not search exchange rates: Kad lists them instead.

Ask, for example: "Which cost centers sit under Group IT?", "Who is the budget holder of FR-TRV-600?", "How many working days are there in France in 2026?", "Which dimensions are required?" or "What is the USD rate for 2026?" Assistants connected over MCP get the same answers.

Totals come from Kadratis. A list of budget lines or contracts comes with the total of every matching record, all pages included, and with the total of the records listed, so Kad quotes one of them or asks Kadratis for a sum. Asked for the ten largest lines and their total, Kad gives the total of those ten. When your data access is limited to some companies or cost centers, Kad says that its counts and totals cover your access, and does not present them as the total of the workspace.

When Kad tries a filter value that a field does not take, for example "active" for the status of an application (which is "enabled" or "disabled"), Kadratis refuses it and lists the accepted values. Kad then searches again with the right value.

When you ask for a change your role does not allow, Kad names the right you are missing, for example "you need the right to modify projects". An administrator adds it to your role.

Kad makes only the changes your screens let you make, on the records you can reach there. With the right to contribute to projects, you can have Kad change the purpose of a project and its links to applications, assets and budget lines. The right to contribute to requests covers the same links on requests. A Business Contributor makes these changes only on the projects and requests they take part in. Linking a record needs the right to read its module. Renaming a project or changing its other fields needs the right to modify projects, as on the project page.

With the Contributor level on incidents, Kad logs an incident, changes its fields, moves its status forward up to Closed, adds notes to the journal and links assets, applications and technologies. With the Member level on tasks, it creates a follow-up task on an incident. With the Administrator level on incidents, it reopens or cancels an incident, with a reason, and turns its restriction on or off. Closing freezes the incident. Deleting it stays on its screen. See Ask the assistant.

With the Member level on infrastructure, Kad creates a technology, adopts it from the Kadratis catalog or completes a technology typed by hand. It changes its fields, adds versions and changes their dates, links it to applications and servers, adds or removes its owners, support contacts and web links, and turns its vulnerability alerts on or off, for the whole technology or one version. Linking an application also needs the right to read applications. Adding a support contact needs the right to read contacts. Ask Kad to undo a change it made in the conversation: it prepares a preview that puts things back. Creating or adopting a technology cannot be undone, and a version added along the way stays. Deleting a technology or a version, attachments, vulnerability identifiers and running the vulnerability matching again (Refresh) stay on the screens. See Ask the assistant.

With the same level, when vulnerabilities are followed, Kad changes the state of vulnerabilities to To review, Handled or Not affected, with a note if you give one. Name a CVE to change all its rows, add a technology or a technology and a version to narrow it, or describe the rows with the filters of the page. One change covers 5,000 rows at most, as on the page. The preview says how many rows and CVE change. Accepted is not a state: Kad offers Handled with a note, or Not affected. Ask Kad to undo the change in the conversation. Ask, for example: "Mark CVE-2024-3094 as not affected on TEC-3, the package is not installed". See Ask the assistant.

When Kad creates a budget line, it asks whether the line is OPEX or CAPEX if you did not say. It does not change the type of an existing line: use Change type... in the line's workspace (see Changing the type). Kad sets the contract of OPEX lines only. A CAPEX line reaches its contracts through the links of its Relations tab.

Deleting a record¶

Kad deletes one record at a time: a task, a contact, a location, a document, a budget line (OPEX or CAPEX), a connection or an interface. Each deletion comes as a preview that you approve. A deletion cannot be undone.

The preview lists what goes with the record, such as comments, time entries, attachments, versions and links. It names only what you may read and counts the rest as other links. If the record changes between the preview and your approval, Kadratis refuses the deletion. Ask Kad again to get a new preview.

Only one deletion waits for approval at a time in a conversation. When you ask to delete several records, Kad prepares them one after the other.

Kad needs the level the screen asks for:

Record	Level
Task	Tasks administrator
Contact	Contacts administrator
Location	Locations member
Document	Knowledge administrator, with the right to write in its library
Budget line	Budget administrator
Connection	Infrastructure member
Interface	Applications administrator

When your data access is limited to some companies or cost centers, Kad deletes no contacts and no budget lines, as on the screens.

Kad refuses what the screens refuse:

A task that a request names as its origin
A budget line with amounts in a frozen column
An interface with bindings, unless you ask to delete its bindings with it
A document embedded in a record. It goes when its record is deleted.

Other records, such as applications, assets, contracts, projects, requests, incidents and technologies, are deleted on their screens. Kad says so and names the screen. Assistants connected over MCP read only.

What Kad does not read¶

Kad does not read:

The audit log (Admin › Audit log)
AI agents

When a question touches one of these, Kad says that it cannot read it and names the screen that shows it, as your menus name it in the language of your profile. A search without a result tells Kad what it did not search, so the answer does not claim that the record is missing. The same goes for a field: the context of a record carries its main fields, and Kad reads the full record before saying that a field is empty. Assistants connected over MCP get the same answers.

Artifacts and previews¶

Some answers come with extra material that does not fit naturally inside the chat thread. Kadratis calls these artifacts.

Common cases:

A long block of text or markdown that Kad prepared for you
A side-by-side Before / After diff of a record Kad wants to update
A draft import or change set that requires your sign-off

Artifacts open in a side panel on the right of the workspace. The panel can be toggled by clicking the Artifacts tab button on the right edge of the screen.

The panel auto-opens when:

A long preview arrives during a streaming response
A pending preview needs your decision (these always pop the panel open, since you must act on them)

For pending change previews, the panel offers two buttons:

Approve: confirms the change and lets Kad apply it
Reject: cancels the change. Kad says the change was rejected and prepares nothing in its place. When your request asked for several changes, it goes on with the others.

A preview, and the message Kad writes after you approve or reject it, are in the language of your profile, with the field names and values the screens use. A budget line asked for with an amount comes as one preview: one approval creates the line, its budget version for the year and the amount.

When you ask for a change Kad cannot make, such as deleting a project or running the vulnerability matching again, it says so and names the screen where you can make it. It prepares no other change in its place.

Nothing that mutates Kadratis data is applied silently. The preview is the gate.

Usage indicators¶

Above the composer, a small indicator shows how much of the included monthly volume remains:

Built-in usage: when Kad is running on the Kadratis included model rather than one of your organization's own models, this shows how many messages remain in the current month and the date the quota resets. An administrator has confirmed that model beforehand (see Kad Settings). When the limit is reached, the composer is disabled and a helper text invites administrators to switch to a model of their own; see AI models.

People who can open Admin > Kad also see a thin bar under the composer, once the conversation has at least one exchange. Volume processed adds the input and output tokens of every message in this conversation. Each question resends the earlier messages, so the volume grows faster than the number of turns. Context of the last request is the input of the latest question. The bar also shows the input tokens and the output tokens of the conversation.

Tips¶

Use prefixes for precision: @T-, @DOC-, @PRJ-, @REQ-, @INC-, @TEC- map directly to native Kadratis references. They are the fastest way to point Kad at a specific record and they survive copy-paste because they look identical to what you see elsewhere in the application.
Start a new conversation per topic: keeping unrelated questions in separate conversations makes the context window smaller, the answers faster, and the token bill lower. The conversation list is grouped by date so you can find them again easily.
Approve and reject deliberately: previews are the only thing standing between Kad and your live data. Take the extra second to read the diff before clicking Approve.
Stop instead of waiting: if Kad heads down the wrong path mid-stream, hit the Stop button rather than waiting for it to finish. You will save tokens, and your follow-up message can correct the course.
Drop images directly: dragging a screenshot onto the composer is faster than the file picker, and pasting from the clipboard works too. Use it when describing a UI issue or asking Kad to read a chart.