CHAT

CHAT

Questions and answers in natural language with the model you choose, with web search, attachments and diagrams drawn on the page.
INDEXES

INDEXES

Your documents, organized by topic and shared with colleagues. The chat answers citing the source, page included.
IMAGES

IMAGES

Generate an image from a description, or attach a photo or a diagram and ask questions about what it contains.
AGENTS

AGENTS

Define an agent once, launch it as often as you need and follow the work in the Console. AI agents for business →
The menu, in short. On the left you find New Chat and Chats (the conversations), Indexes (the documents), Agents (Configurations and Executions), Semantic Search to find a passage across all indexes without opening a chat, Summaries for repeatable summaries and questionnaires, Templates for reusable texts, API Keys to integrate AIDeskPro into your systems and Connector to link external sources.

The Chat

After logging in you land on New Chat. Before writing you can decide three things: which model to talk to, on which documents and with which tools. Or just write: the default model answers right away.

Choosing model, index and tools

  1. In the CHAT ENGINE bar click the pencil to change model. In the dialog you can also tick Set as my default and keep it for future chats.
  2. Next to it, NO INDEX opens the list of Indexes: pick one or more, and the chat will answer only from those documents.
  3. The gear icon at the top right opens the tools: Internet for web search, Image generation for images, Documents for attachments.
  4. The button next to the message box attaches a file or an image to the conversation.
  5. Every chat gets an automatic title; from the bar you can share it, print it or download it.
New chat with the tools panel open
The New Chat page with the tools panel: Internet, Image generation, Documents.

A work question

The plain chat is fine for everything that doesn’t need your documents: rewriting an email, translating, summarizing a pasted text, drafting an outline.

Chat: rewriting an email in a formal tone with Gemini 4 Argon
Rewriting an email in a formal tone. At the top the model used, Google Gemini 4 Argon, and no index.

Turn on Internet among the tools and the model searches the web before answering. The answer lists the sources with their links, so you can check where each piece of information comes from.

Chat with Internet search on and cited web sources
A question about a recent event: the answer cites the web pages it found.

Diagrams

If you ask for a flowchart, an entity-relationship schema or a sequence diagram, the chat doesn’t give you code back: it draws the diagram on the page. It works well for explaining a process, a data model or a protocol.

Flowchart of an offer approval process drawn in the chat
The approval flow of a sales offer, asked as a compact left-to-right flowchart.
Sequence diagram of the TCP three-way handshake drawn in the chat
A sequence diagram: the TCP three-way handshake between client and server.
Entity-relationship schema of customers, orders and products drawn in the chat
The ER schema of a small management system, with keys and cardinalities.

Asking questions about an image

With a multimodal model you can attach an image and ask what it contains. Here the legend of a remote control’s buttons exists only in the figure: the model reads it and answers.

Chat with an attached image and the answer describing the remote control buttons
The image attached from the button next to the message, and the answer that reads the figure.

Indexes

An Index is a collection of indexed documents, that is, documents made queryable by a chat. It can hold one file or thousands. Our advice is one per topic: company policies, offers, product manuals, a customer’s contracts.

List of Indexes
The Indexes page: your indexes and the ones shared with you.

Creating an Index

  1. From the Indexes menu, click New index.
  2. Give it a name and choose the type: ExtendedSummary for documents (the recommended one), Composed for an index made of other indexes, Chat if it only holds instructions.
  3. Save: the Documents, Share, Advanced settings and Metadata tabs appear.
Form for creating a new Index
Name, type and embedding model: the rest is configured after saving.

Uploading documents

  1. In the Documents tab click Add and choose the file. PDF, Office, text, HTML, images and many more formats are accepted; with Add folder or Add ZIP you upload a whole folder.
  2. The title is proposed from the file name: change it to something readable, it is what the chat will cite as a source.
  3. Indexing starts by itself. When the status is OK the document can be queried; Update documents refreshes the list.
Documents tab of an Index with three policies in OK status
Three company policies indexed and ready.
Share tab of an Index
The Share tab: who can read and who can edit.

Sharing

  1. In the Share tab add the colleagues or groups to share the Index with.
  2. Choose whether they can only query it (read only) or also add documents.
  3. Without shares the Index stays private, visible only to you.

Querying the documents

Open a chat, pick the Index from NO INDEX and ask. The answer is built only on the documents, and at the bottom you find Sources: every passage used, with the document, the page and the button to download it.

Chat on an Index: answer about backup and access policies
Four questions about the policies, one answer with the numbers taken from the documents.
List of the sources of an answer, with document and page
The sources: document, page, the cited passage and the download.

Generating images

With the Image generation tool on, describe in the chat the image you want: style, subject, colors, format. The image appears in the answer and can be downloaded with the icon at the top left. Images are generated by Nano Banana, the engine integrated in AIDeskPro. To modify an existing image, attach it and describe the change.

Image generated in the chat: a robot delivers a letter in an office
A newsletter cover, generated from a two-line description.

Agents

A chat answers; an agent works: it receives a mandate, reads the documents, writes files, checks and delivers a result. In AIDeskPro an agent is defined once as a Configuration and used as often as needed as an Execution. Whoever writes the Configuration knows the process; whoever launches the Execution only needs to know what they want to get.

The example in this guide is an agent that fills in the security questionnaires customers send to their suppliers, using the company policies as its source.

The Configuration

  1. From Agents → Configurations click New agent configuration.
  2. Give it a name, a description and write the System prompt: who the agent is, which resources it uses, in which phases it works, what it must never do.
  3. Save. The Integrated functionalities (web search, images), Files (files always available in the workspace, for example a template), Resources and Outputs sections appear.
Agent configuration: name, model, description and system prompt
The Configuration: name, model, description and the prompt with mandate, resources, phases and rules.
Resources section of a Configuration with two Index slots
Two slots of type Index: one required, one optional, both chosen at launch.

Resources

  1. In Resources click Add and choose the type: Index (documents), Sub Agent (another agent), Database (via MCP JDBC) or External (an MCP server).
  2. Give the slot a name and a description: the prompt refers to it by name, here policy and risposte-precedenti.
  3. Set now fixes the resource in the Configuration, the same for everyone; Request at execution lets whoever launches choose it.
  4. Required: without that resource the agent doesn’t start.

The Execution

  1. From the Configuration click New execution, or from Agents → Executions click New agent execution and pick the Configuration.
  2. Write the title and the starting command: the mandate, as you would give it to a colleague. Choose the model.
  3. Save: for each Request at execution slot choose the Index with Select. The warning triangle marks required slots still empty.
  4. In Files you can upload the documents of the case. Then Start.
New Execution: title, model and starting command
The Execution: title, model and the starting command.
Choosing the Index for a slot at launch time
The policy slot asks for an Index at launch: here Policy aziendali.

The Console

As soon as it starts, the Execution shows the Stop and Console buttons. The Console is where you follow the work: on the left the conversation with the agent, on the right the Workspace files with a preview. A well-written agent, when something is missing, asks before starting, with a closed question.

Console of an Execution: the agent asks for the missing file with a multiple-choice question
The questionnaire isn’t in the workspace: the agent asks for it, with three possible answers, before beginning.
Console with the agent's steps and the workspace files
The agent at work: the steps in the conversation, the files it writes in the Workspace, the preview on the right.
Console at the end of the job: the completed questionnaire and the summary of questions to decide
At the end: the completed questionnaire in the Workspace and, in the chat, the four questions without a source marked DA DECIDERE for the IT manager.

Following, stopping, resuming

  1. Upload files to the Workspace with the icon at the top right of the Console; the agent sees them immediately.
  2. Answer the agent’s questions from the conversation, as in a chat.
  3. Stop halts the execution, Resume picks it up where it was: useful to fix a wrong index without starting over.
  4. The files produced stay in the Workspace: open them, correct them, download them. No result leaves the company without a person deciding it.
How to write a good agent, with phases, file-based memory and independent checks, is in How to write an agent . What changes in a company when you move from chat to agents is in Enterprise agentic systems .

For administrators

Two pages users don’t see, but that decide what AI is allowed to do in the company.

Model Permissions page: models, publisher and visibility
Model Access: which models, for which groups.

Model Access

  1. Every model has a visibility: public for everyone, or restricted to user groups.
  2. Models with an endpoint outside the European Union are labeled No EU: they are used only when someone deliberately picks them, never as defaults.
  3. Groups are managed in UserGroups; users come from the company single sign-on.

Cost dashboard

  1. Dashboards shows daily costs, tokens per model and spending per user label.
  2. Pick the period and the user group to see who uses what.
  3. A cost that suddenly rises is often the first sign of anomalous use: thresholds and alerts are there to see it before the invoice.
Cost dashboard with the daily trend
The default dashboard: daily costs and costs per model.

Need support?

Read the FAQ →