vragen.ai with agent mode: one question, several reasoning steps, a better answer

With vragen.ai, visitors ask their question in their own words. They get a reliable answer right away, based on your own content. With sources cited, so it is always clear where the information came from.

Until recently this worked in a fixed order:

1. Retrieval: vragen.ai pulls relevant passages from your knowledge base.

2. Generation: from those, it formulates an answer.

That works well, but it has a limit: there is only one round of searching. Which means valuable context sometimes stays just out of view.

Agent mode changes that, fundamentally.

What is agent mode?

Agent mode (agentic AI) means vragen.ai no longer takes one linear step from search to answer, but can take several search and reasoning steps on its own to arrive at the best answer.

• The agent looks up relevant information.

• Does it notice something missing, or a smarter path?

• Then it searches further.

• For as long as it takes to get the context right.

This way of working fits a broader development in AI: agentic RAG, where systems search and reason iteratively instead of trying to retrieve everything in one go.

What are the advantages?

1) A smarter retrieval process

An agent doesn't only search for what the question says literally, it builds on what it finds along the way. So it retrieves broader and more relevant passages.

An example:

• A visitor on a patient platform asks about peer support.

• The AI finds a passage in which someone got in touch with others through a drop-in center.

• The agent decides there is something here, and searches on for similar experiences around the drop-in center.

• That builds extra context before the answer is given.

The result: less missed information, and a more complete answer.

2) More room to ask follow-up questions

Instead of only broadcasting, the agent becomes someone to talk to.

• When something isn't clear, it asks a question of its own.

• It understands follow-up questions better, because it built the context itself.

It feels more like a sounding board than a Q&A.

3) Better marking of the sources used

Agent mode also sharpens the way sources are referenced.

• The text that was used is clearly marked on the page.

• You see exactly which passage was used, and why.

• That gives end users confidence.

Evaluations of agentic RAG point to this kind of span-level citation as an important gain in quality.

Example: what can an agent do that used to be difficult?

Question: "What is the phone number of the Product Owner of Team Delta?"

An agent can now work through this on its own:

1. Find out who the Product Owner of Team Delta is.

2. Then search specifically for contact details.

3. And only then give the answer, with the source cited.

So you no longer have to work out the intermediate steps yourself or ask several separate questions. The agent does that for you.

Several agents, each with its own role and tone

One important advantage is that you can run several agents side by side, each with its own task and tone.

For each agent you set:

• What it is meant for

• Which tone you want (formal, friendly, concise, and so on)

• The length of the answers

• Which documents from the knowledge base it may use

• How much room it gets to reason (the reasoning effort)

Creating one is straightforward: you pick a type, give an instruction, and vragen.ai automatically writes a matching prompt. You can always review or adjust it yourself.

Tip. Working with different audiences? Then make a separate agent per audience, tuned to the right tone, with sources that match what that group needs.

What does the user notice?

For end users, how the answer comes about becomes more transparent.

• You see which steps the AI takes to reach the answer.

• Follow-up questions are part of it (agents work iteratively).

• Source references are clearer and marked more precisely.

That makes the answer better, and checkable.

Why this opens up new uses

Because agents can search, probe and reason on their own, uses become possible that used to take a lot of manual work. Think of:

• Smart customer service
  The agent searches through FAQs, policy and cases. And asks clarifying questions itself.

• An onboarding agent
  Combines handbooks, project documents and role information into a clear starting point for new colleagues.

• A policy or legal assistant
  Connects different versions and flags gaps or ambiguities.

• Product or service advice
  Asks for the right context first, searches on from there, and only then gives fitting advice.

• Live data connections
  Through an MCP connection the agent can also retrieve external, current data. Stock levels, statuses or CRM information, for example.

  👉 We'll publish a separate article soon on what MCP is exactly and how to use it with vragen.ai.

Agent mode is not an extra button. It is a way to use vragen.ai much more as an intelligent assistant.

In short

Agent mode turns vragen.ai from an AI search engine that searches once into a digital partner that keeps searching, asks follow-up questions and builds up context.

You get:

• Smarter, iterative retrieval

• More active sparring and better follow-up questions

• More precise source citation

• And above all, room for uses that were not possible before