Asking an enumerating question

User

This guide shows how to ask a question whose answer is a complete list or a count across your entire knowledge base — for example “Which customers have a data processing agreement?”. For that you need research mode. → Research mode

Prerequisite

Research mode has to be enabled in your instance. If you can’t find the toggle, contact your administrator. → Enabling research mode

Step 1 — switch on research mode

Open Chat and switch on the Research chip (magnifying glass icon) in the composer — that’s the research mode toggle. It sits next to the Private and Knowledge base toggles.

Don’t confuse it with the Web chip (globe icon), if you see it: that one researches the web instead of your knowledge base. For an enumerating question about your own records, Research is the right chip; both at once is not possible. → Research or Web?

The toggle is off by default and applies only to the current tab. Switch it on before you send the question — it does not apply retroactively to a question you’ve already asked.

The toggle applies to exactly one message. It resets automatically on send. So switch it on again for every further enumerating question. Forget that and you get an ordinary — and, for an enumerating question, incomplete — answer back, with no error message.

Step 2 — phrase the question so that it enumerates

Phrase the question so it’s clear you want all hits, not the best ones:

Good: “List all customers for whom a data processing agreement exists.”
Good: “How many projects use PostgreSQL? Name the projects.”
Less good: “Tell me about our data processing agreements.” — that’s not an enumerating question; normal chat is faster and more appropriate here.

State the criterion as concretely as possible. The run sifts every document against exactly this question — the clearer it is, the more useful the report. A narrowly framed question is also cheaper than a broad one, because only the matches get evaluated.

Send the question.

Step 3 — expect the placeholder notice, not the answer

The chat turn ends within a few seconds with the notice: ”🔍 The background research across the knowledge base is running. The report will be delivered to this conversation automatically as soon as it is finished.” This is not the answer, and not an error either.

You now need to do nothing: no waiting, no follow-up question, no reload. Carry on working normally — you can ask other questions or switch conversations.

You can tell the run is still going from the composer: the magnifying glass turns into a spinner showing the current phase (Planning → Selecting → PII check → Evaluating → Summarizing → Composing report). The display stays put across conversation switches and page reloads.

Step 4 — receive the report

After a few minutes the report appears as a new assistant message:

If you are still in the same conversation, it shows up there by itself.
If you have switched conversations, a badge appears on that conversation in the list. Click it to jump to the report.

Before reusing the report, it pays to look at its start and its end: the coverage header says how many documents were searched and with which search terms, and the appendix “Cited sources (N)” lists every document a finding came from. If the number of matches does not match your expectation, the search terms are usually to blame — rephrase the question in your documents’ vocabulary. → How a report is structured

Step 5 — reuse the result

The report is a normal assistant message, so you can reuse it like any other answer:

Ask follow-ups — put follow-up questions about the report in the same conversation. You do not need research mode for those; normal chat answers them from the report. For a new enumerating question you do.
Export it as Word or PDF. → Exporting an answer
Save it to the knowledge base if the result is needed permanently. → Saving an answer to the knowledge base

If no report arrives

Did you just get a fast, ordinary answer? Then the toggle wasn’t on when you sent — most often because it reset itself after your previous question. Switch it on and ask again.
Is the desktop app running? On desktop the research only runs while the app is open. If it was closed in the meantime, the run is lost; on the next start the conversation reports “The application was shut down while the run was in progress.” Ask the question again.
Did “Research (aborted)” or “Research failed” arrive instead of the report? Then the run ended early; the cause is under “Reason:”, and whatever was finished by then follows below. → If the run does not end normally If the reason concerns configuration (a missing chat provider or the PII service, say), only an administrator can help. → Enabling research mode
Did the ordinary answer start with “ℹ️ A background research run is already in progress. This question was therefore answered normally.”? Then one of your research runs was still going — Research or Web, there is only ever one at a time. Wait for its report (the spinner in the composer shows it is still running), then switch the chip on again and ask again.
Did “⚠️ The research could not start: its research agent is disabled or was deleted. …” arrive? Then the agent that carries out the research is switched off on the Agents page. Only your administrator can fix that.