Vragen.ai with agent mode: one question, multiple thought steps, better answer
With agent mode, Vragen.ai becomes much smarter. Instead of searching once and then answering, the AI now gets to continue searching, probing further, and building context on its own.
Joris Meijer
Co-founder, AI Lead
With Vragen.ai, visitors can ask their questions in their own words. They instantly receive a reliable answer based on your own content, with a source reference, so it's always clear where the information comes from.
Until recently, this worked in a fixed sequence:
Retrieval – Vragen.ai retrieves relevant pieces from your knowledge base.
Generation – based on that, it formulates an answer.
This works well but has a limitation: there is only one search round. This sometimes causes valuable context to remain just out of sight.
With agent mode, this changes. Fundamentally.
What is agent mode?
Agent mode (agentic AI) means that Vragen.ai no longer performs one linear step “search → answer,” but independently takes multiple search and thinking steps to arrive at the best answer.
The agent searches for relevant information.
If it notices something is missing or another path seems smarter?
Then it continues searching.
Until the context is correct.
This way of working aligns with a broader development in AI: agentic RAG. Systems search and reason iteratively, instead of trying to retrieve everything at once.
What are the advantages?
1) Smarter retrieval process
An agent not only searches for what is literally in the question but builds on what it finds in the meantime. This allows it to retrieve broader and more relevant fragments.
Example:
A visitor asks a question about peer support on a patient platform.
The AI finds a fragment where someone came into contact with others through the drop-in house.
The agent decides: there's something here. And searches further for similar experiences around the drop-in house.
This way, additional context is created before the answer is given.
The result: less missed information and a more complete answer.
2) More room for follow-up questions
Instead of just broadcasting, the agent becomes a conversation partner.
If something is unclear, it asks a counter-question.
Follow-up questions are better understood because it has built the context itself.
You get more of a sparring partner experience than a Q&A.
3) Better marking of used sources
Agent mode also ensures sharper references in the source.
The used text is clearly marked on the page.
You see exactly which fragment is used and why.
This gives end-users confidence.
Evaluations around agentic RAG refer to such ‘span-level citations’ as significant quality improvements.
Example: What can an agent do that was previously difficult?
Question: “What is the phone number of the Product Owner of Team Delta?”
An agent can now independently:
Determine who the Product Owner of Team Delta is.
Then specifically search for contact details.
And only then provide the answer, with source reference.
You no longer need to think of intermediate steps or ask multiple separate questions. The agent does that for you.
Multiple agents: each with its own role and tone
An important advantage is that you can engage multiple agents side by side, each with its own task and tone.
Per agent, you set:
The purpose
The desired tone (formal, friendly, concise…)
The length of the answers
Which documents from the knowledge base may be used
How much reasoning effort it is allowed (the “reasoning effort”)
The setup is straightforward: you choose a type, give an instruction, and Vragen.ai automatically creates an appropriate prompt. You can always review or adjust it yourself.
Tip. Are you working with different target groups? Then create a separate agent for each group. Tailored to the right tone, with sources that match what that group needs.
What does the user notice?
For end-users, it's more transparent how the answer comes about.
You see which steps the AI takes to come to the answer.
Follow-up questions are part of it (agents work iteratively).
Source references are clearer and more tightly marked
This makes the answer not only better but also verifiable.
Why this allows for new applications
Because agents can independently search, ask questions, and reason, applications that previously required a lot of manual work arise. Consider:
Smart customer service
The agent searches through FAQs, policies, and cases. And asks clarifying questions itself.Onboarding agent
Combines manuals, project documents, and role information into a clear starting point for new colleagues.Policy or legal assistant
Connects different versions and signals gaps or ambiguities.Product or service advice
First solicits the right context, searches thoroughly, and then provides a suitable recommendation.Live data connections
Through an MCP connection, the agent can also fetch external, current data. Such as inventory, statuses, or CRM information.
👉 We will soon publish a separate article about what MCP is exactly and how you can use it in combination with Vragen.ai.
In short: agent mode is not an “extra button” but a way to use Vragen.ai much more as an intelligent assistant.
Summary
With agent mode, Vragen.ai transforms from an AI search engine that searches once, into a digital partner that continues searching, asking questions, and building context.
You get:
Smarter, iterative retrieval
More active sparring and better follow-up questions
More precise source referencing
And above all: room for new applications that were previously not possible
