Review a document with AI, no setup

A situation every business knows

A contract is waiting for signature, and nobody has read all of it. A call for tenders has arrived, and you want to know whether you even have the required certificates. An official notice lands on your desk, and the deadline that matters is somewhere on page four. Two quotes sit side by side, and the differences are hidden in the terms and conditions, not in the price.

Document chat exists for exactly these cases. You upload the file and ask your question. There is nothing to set up and nothing to prepare: document chat is available to every organization from the start, and every member may use it.

Your first answer

1. Open Document chat in the sidebar.
2. Click New analysis. The analysis is created immediately and opens.
3. Drag your file into the upload area, or click it to choose the file. Wait until the progress bar has finished.
4. Read the Overview the AI creates on its own the first time you open the document. Below it are one to three starter questions that fit the type of document. One click sends the question.
5. Or type your own question into the input field at the bottom and press Enter.

Questions that work in practice

Document chat answers from the content of the uploaded file, not from general world knowledge. So ask the way you would ask a colleague who has just read the paper:

• Contract before signing: "Which notice periods and terms apply, and which clauses bind us beyond the end of the contract?"
• Checking a tender: "List every required certificate, reference and deadline, with the passage where it appears."
• Understanding an official notice: "What is required of us here, by when, and what are the legal consequences if we do nothing?"
• Comparing quotes: "Apart from price, where do the two quotes differ: scope of services, warranty, payment terms?"

For the last case, upload both quotes into the same analysis. Only when the documents sit together in one analysis can the AI compare them against each other.

Every statement with its source

Answers contain clickable source references. A click opens the original document at exactly the page the statement comes from. That way you check an answer instead of trusting it. With a contract or an official notice, you should do that every time.

Keep what you want to hold on to:

1. Click Save at the bottom right of the answer. The button is always there, so you never have to look for it. The answer is stored as a finding on the Saved tab.
2. Click Export at the top right to output your saved findings together with the conversation: as Markdown (.md), PDF (.pdf) or Word (.docx), or with Copy to clipboard.

The conversation stays until you delete the analysis. If you open it again next week, the conversation is still there.

Limits to know in advance

One restriction applies to the formats: only with PDF does a click on a source reference jump straight to the right page. With Word, text and Markdown, the same click opens the document, but without jumping to the passage.

The page limit is the one you are most likely to hit: a call for tenders or a bill of quantities often runs well beyond 50 pages. Put documents like that into a knowledge base instead, which has no page limit, and query them from there.

Scanned paper does not work here. Document chat reads the text contained in the file. A PDF made only of images, such as a photographed or scanned sheet without a text layer, is accepted, but it gets the Scan label and the warning No readable text detected: the AI cannot read anything from it. It still takes up one of your five slots, so it is best to delete a document marked like this right away. Scans belong in a knowledge base, which comes with text recognition.

Two more things you do not need to look for: switching models is deliberately not offered here, because a model suited to reading documents always runs. And your analyses belong to you alone. Other members of your organization cannot see them, whatever their role.

Document chat or knowledge base?

This is the most common mix-up, and it costs time in both directions. The decision comes down to a single question: is this a one-off case or a standing collection?

A one-off case is this one contract, this one tender, this one notice. You review it, you make a decision, and then the matter is closed. That is what document chat is built for: no setup, no permissions, nothing you have to maintain afterwards. Putting a review like that into a knowledge base means building structure for something you will never need again, and it loads other parties' draft contracts into a collection the whole team searches.

A standing collection is what you need again and again: price lists, work instructions, bylaws, manuals, your set of standard contracts. For that, use the knowledge base. It is permanent, accessible to the whole team, has no 50-page limit, and it can read scanned paper, which document chat cannot. If you upload the same file again every day, you are working against the tool.

If you realize along the way that the one-off case has become a standing collection after all, you do not have to start over: an existing analysis can be moved into a knowledge base.

When a question becomes a real review

Sometimes the answer in the chat is not enough. You do not need a conversation but a result you can pass on: a risk assessment of the contract as a Word file, a tender checklist as a PDF for management.

That is what the agents are for. You describe your assignment in plain language, and the system passes it to the right specialist; for a document review, that is the analysis agent. You do not have to pick anyone or remember who is responsible for what. Unlike document chat, the agents deliver a finished file: a PDF, a Word document or a calendar entry you can take over directly.

How to make the switch:

1. Open Agents in the sidebar.
2. Describe the assignment in the text field, as specifically as possible, and upload the file again. Input files can be up to 50 MB here, much larger than in document chat.
3. You will not find a knowledge base picker here, and you do not need one: the agents search every knowledge base you have access to on their own, so your own standards are set against the other party's paper without any extra step.
4. Send the assignment with the arrow button to the right of the input field (Send message) and approve at the checkpoints.

These checkpoints are where you stay in control. The agent first presents its plan, asks when something is unclear and, for elaborate assignments, gets your approval before it continues. For every assignment you can also open a traceability view that records in plain language what the agent did, which sources it read and which approvals you gave. Agents reply in the language of your assignment, or in your interface language if that is unclear.

One thing to know in advance: the agents are not active from the start. Datenmassiv AI switches them on for your organization: a short conversation, no application form. Document chat and knowledge bases, on the other hand, are ready immediately, so you can start reviewing while the activation is still under way.

Read more

• Document chat: every step, limit and error message in detail.
• Agents: the agents, their tools and permissions.
• Knowledge bases: standing collections instead of one-off cases.

Next: AI meeting minutes from a recording, how a recording becomes a usable draft without rework.