AI Chat
AI chat is the heart of the Queria experience. Through an intuitive conversational interface, you can query your organization's entire documentary heritage, get circumstantiated answers with citations to sources and generate new documents.
How to ask effective questions
The answer quality largely depends on how you phrase the question. Queria's Cog-RAG architecture is designed to understand natural language, but some practices noticeably improve results.
Specific vs generic questions
| Less effective | More effective |
|---|---|
| "Tell me about contracts" | "What's the duration and value of the maintenance contract with supplier Rossi?" |
| "What does the regulation say?" | "What obligations does current regulation provide for special waste disposal?" |
| "Summarize the balance sheet" | "What was the EBITDA trend in 2025 compared to 2024?" |
Supported question types
- Factual: "When does contract X expire?" -- pointed answers from a single document.
- Aggregative: "How many active contracts do we have with expiration in 2026?" -- data collection from multiple documents.
- Comparative: "What are the differences between policy A and policy B?" -- side-by-side analysis.
- Exploratory: "What do we know about project Alpha?" -- broad search with progressive synthesis.
- Generation: "Generate minutes of the meeting on February 15" -- new document creation (see Document Generation).
Anatomy of the answer
Every assistant answer is composed of several elements that help you evaluate accuracy and origin.
Text with inline citations
The body contains numbered references in square brackets: [1], [2], [3]. Each number uniquely identifies a source. The same answer can draw on multiple documents and sources, weaving the information coherently.
Example:
The contract provides a 2% penalty for each week of delay [1], with a maximum of 10% of total value [1]. This clause is in line with the regulations in force on the matter [2].
Source badges
At the bottom of the answer, a dedicated area shows the used sources as colored badges:
| Color | Source type |
|---|---|
| Blue | Uploaded company documents |
| Green | Documents from the company Knowledge Base |
| Amber | Certified external legal sources |
| Orange | Certified external food sources |
| Teal | Certified external chemical sources |
| Purple | Certified external pharmaceutical sources |
| Gray | Indexed web pages |
Clicking a badge opens the original document passage, allowing direct source verification.
Reasoning panel
Clicking "Show reasoning" expands a panel that reveals the assistant's decision process:
- Question analysis: how the Planner interpreted your request.
- Search strategy: whether the question was decomposed into sub-queries (sequential, parallel or comparative).
- Sources consulted: which collections and sources were queried.
- Result selection: how many documents were found, filtered and re-ranked.
- Confidence level: how confident the assistant is in the provided answer.
This panel is particularly useful to understand why the assistant gave a certain answer and to refine subsequent questions.
Follow-up suggestions
After each answer, the assistant proposes 2-3 related questions as clickable buttons. Suggestions are personalized at three levels:
- From the current answer: specific aspects to deepen
- From the conversation: linked to facts emerged in previous turns
- From your profile: at most 1 suggestion linked to your interests, only if pertinent to the current topic
E.g., if you work with insurance contracts and ask about a liability clause, a follow-up might suggest comparing it with IVASS regulations you often consult.
Actions on answers
Hovering over an assistant answer, several actions appear:
- Regenerate answer: re-process the answer to the same question with a new generation.
- Continue generating: if the answer was truncated for length, continue generation from where it stopped.
- Copy answer: copy the text to clipboard, formatted in Markdown.
- Export as document: download the answer as a file (see Document Generation).
- Positive/negative feedback: thumbs up or down to help improve the system over time.
Temporary documents in chat
You can attach documents directly in the chat using the paperclip icon or drag & drop. These documents are temporary: they remain available only for the current conversation and are automatically removed at closure.
This feature is ideal for:
- Analyzing a received document on the fly.
- Comparing a new contract with those already in the system.
- Extracting specific information from a file without adding it to permanent storage.
Temporary vs permanent documents
Documents attached in chat are not saved to the Knowledge Base and won't be available in future conversations. To make a document permanent, upload it via the Document Management section or add it to the Knowledge Base.
Topic selection
At the top of the chat you can select a topic (category) to narrow the search field. When a topic is selected:
- The assistant searches only among documents associated with that topic.
- AI behavior can be customized per topic (e.g. a "Legal" topic can have specific instructions for legal language).
- Answers will be more pertinent and faster because the system works on a targeted document subset.
To dig deeper, see the Topics & Categories guide.
Smart filters
The filter panel (accessible via the filter icon on the right) lets you further refine the search:
| Filter | Description |
|---|---|
| Document type | Filter by format: PDF, DOCX, XLSX, HTML and more. |
| Date range | Limit the search to documents from a specific period. |
| Topic | Select one or more categories. |
| Entities | Filter by people, organizations or places mentioned in documents. |
| External sources | Enable or disable single categories of certified external sources. |
| Confidentiality | Filter by confidentiality level (public, restricted, confidential). |
Filters can be combined. For example, you can search only PDFs uploaded in 2025 that mention a specific organization.
Conversational memory
Queria's assistant maintains context memory at two levels:
Memory within the conversation
Within a conversation, the assistant remembers everything said. You can reference previous questions and answers without repeating them:
"Go back to point 3 of the previous answer and elaborate." "What is the maximum limit of that penalty?"
The assistant will understand the reference and respond accordingly.
Memory across conversations
Queria stores your preferences and relevant context also across different sessions. If you've been working on a specific subject, the assistant can recall that context in following conversations, making interaction increasingly efficient over time.
Document generation from chat
One of Queria's most powerful features is the ability to generate professional documents directly from chat. You can request:
"Generate explanatory notes based on the balance sheet data I uploaded" "Create meeting minutes from the notes attached"
The assistant will guide you through template selection, data extraction, completion of missing information and DOCX file generation. For all details, see the dedicated Document Generation guide.
In-chat help
If you have questions on how to use the platform, you can ask the assistant directly:
"How can I upload a document from the company network?" "What are topics and how do I use them?"
The assistant can provide operational guidance on Queria use. For more complex matters or dedicated assistance, visit help.queria.pro.
Managing conversations
History
All conversations are saved automatically and visible in the sidebar, sorted by most recent.
Renaming
Click the conversation title to edit it. Use meaningful names to find them easily (e.g. "Rossi contract analysis" or "2025 balance sheet questions").
Deleting
Click the trash icon next to a conversation to remove it. Deletion is permanent.
New conversation vs Continue
Start a new conversation when you completely change topic or want to start without previous context. Continue the same conversation when questions are related, you're deepening a topic or want to reference previous answers.
Tips and best practices
- Start with specific questions: the more precise the question, the more accurate and faster the answer.
- Use filters when you have many documents: narrowing the field improves both relevance and response times.
- Check the citations: always verify sources by clicking on badges, especially for important decisions.
- Leverage reasoning: the reasoning panel helps you check whether the search covered the sources you expected.
- Give feedback: thumbs up and down help the system improve over time.
- Use suggested follow-ups: they often anticipate useful questions you wouldn't have asked.
- Attach documents for quick analysis: temporary documents are perfect for ad-hoc analyses without polluting the archive.
- Select the right topic: if you know which domain your question falls in, select the corresponding topic for more targeted answers.
Queria v3.5.0 -- Canvas-native chat pipeline (DSL)