Datarooms

What is a Dataroom?#

A Dataroom is a secure space where you can store and query uploaded documents and web pages. A Dataroom can contain information on a specific topic or project. The knowledge in a Dataroom can be shared with other users to facilitate collaboration.

Privacy and Access#

By default, new datarooms are private. Only you can see and access them. You can share datarooms with specific users or groups as needed.

Connecting Datarooms to Chats#

Chats are not permanently linked to a single dataroom. Instead, you can dynamically connect one or more datarooms to any chat conversation. This allows you to:

• Query multiple datarooms simultaneously in a single conversation.
• Switch between different dataroom combinations as your questions evolve.
• See which datarooms were used to generate each response.

Agent mode in Dataroom Chats#

Models that support agent mode explore the connected datarooms on their own: they search, list and read files and run several steps. This enables use cases such as exploratory research and multi-step retrieval across many documents. Which models support it is shown as Agent mode for datarooms and files in the model selection.

When to use a Dataroom?#

Use a Dataroom when you want to ask questions about the content of uploaded documents and web pages. It is deliberately avoided to rely on the general knowledge of the Large Language Model to reduce the likelihood of hallucinations.

Targeting specific files#

You can direct AITEZA to focus on a particular file by mentioning its exact name (or a distinctive part of it) in your message. This improves the chance that content from that file is prioritized in the answer.

Examples:

• “In project-plan.pdf, what are the major milestones?”
• “Summarize the pricing section from contract.docx.”
• “What does architecture_overview.md say about microservices?”

Tips:

• Use the displayed file name (case insensitive matching works, but quote or backtick the name for clarity).
• If multiple files share similar names, add extra context (e.g., “the PDF version” or a folder hint) so retrieval is more precise.
• Avoid requesting the entire raw text; ask for summaries or specific information.

Hybrid search capabilities#

Hybrid search combines semantic embeddings with lexical (keyword) matching. This means you can ask questions that rely on specific terms, even if those terms are uncommon. Search results are sorted by relevance, so the most useful matches for your query appear closer to the top.

What hybrid search improves:

• Term-specific questions (“Where is ‘AITEZA’ mentioned?”)
• Clarifying definitions (“Define the term ‘vector index’ as used in the documents.”)
• Narrow queries mixing keywords and intent (“In risk_register.xlsx, list mitigation actions for ‘Data Loss’”)

With models that support agent mode, AITEZA can also run a full-text search over all files, which lists every matching file with the number of matches.

What is limited with models without agent mode:

• Listing every document that contains a term.
• Exact frequency counts across the whole dataroom.

Not provided with any model:

• Full verbatim dumps of file contents.

Precise Spreadsheet Analysis (Excel & CSV)#

Classic language models can struggle with large spreadsheets. Many rows, many numbers, and exact calculations are hard to get right from reading text alone. To solve this, models that support agent mode can run precise data queries directly against Excel and CSV files in your dataroom. This enables reliable filtering, grouping, sums, averages, and comparisons.

Behind the scenes, these evaluations are carried out using SQL, a language designed to query and calculate data from tables precisely. You do not need to know or write SQL yourself: AITEZA automatically translates your question into the appropriate query. Every executed query and other tool call is displayed in the chat in a clear, understandable way, so you can always see exactly how an answer was produced.

Example questions you can ask without knowing any SQL:

• “In sales_2025.xlsx, what is the total revenue per region?”
• “In expenses.csv, list all entries above €10,000, sorted by amount.”
• “Compare the average delivery time between warehouse_a.xlsx and warehouse_b.xlsx.”
• “How many entries in customers.csv have status ‘active’?”

Questions that may not work as expected:#

• Return the full raw content of “document.pdf”.
• How many times does the term ‘AITEZA’ appear across all documents? (only reliable with models that support agent mode)
• What is the average length of the documents that mention the term ‘AITEZA’?

Questions that work in a Dataroom:#

• What is the purpose of ‘AITEZA’? (semantic)
• Where is ‘AITEZA’ mentioned and what is said about it? (hybrid)
• List all documents that mention the term ‘AITEZA’. (with models that support agent mode)
• In project-plan.pdf, what are the timeline milestones? (file targeting)
• Summarize the differences between proposal_v1.docx and proposal_v2.docx. (comparative)
• Compare ‘AITEZA’ with other AI solutions. (semantic)

Dataroom Overview#

On the Datarooms page, each dataroom is shown as a card with its name and quick indicators of its content, including the number of files, images, and websites it contains, and its sharing status. This gives you an overview of a dataroom’s size and content, including how much visual material it holds, without having to open it first.