> ## Documentation Index > Fetch the complete documentation index at: https://help.iteam.works/llms.txt > Use this file to discover all available pages before exploring further. # Discovery — investigation in a loop > In Discovery mode you give a goal instead of a question, and the agent investigates in rounds: it raises hypotheses, queries the data, supports or discards each one, and compiles everything into a report with charts. In a normal chat you ask a **question** and get an answer. In **Discovery mode** you give a **goal** — and the agent investigates in rounds until it closes, on its own. A discovery in progress ## Turning it on Pick an agent that has access to the data you want to investigate. It sits in the message bar. A notice confirms it: *your next message becomes the goal*. "Find where the accounts payable cycle loses money" works better than "how much did we spend last month". Ask explicitly for what you want back: **tested hypotheses**, **gaps and opportunities**, **impact in currency**, and a **chart**. The agent delivers what you ask for — what you don't ask for tends to come back as prose. ## What happens during the investigation The agent builds a plan and opens several analysis **fronts**. Each round it queries the data, concludes, and uses what it found to decide the next round. Shows where it stands — for example, *3 of 12 analyses*. You can leave the screen and come back: the investigation keeps going. The queue of questions it still plans to investigate. This is where the iterative nature becomes visible — it's not a single answer. Every hypothesis is checked against real data. When one doesn't hold, it **says so** instead of forcing a conclusion. ## What comes out Hypothesis, cases, value, and verdict — side by side. Built on the spot from what was found. Sized in currency, so you can prioritize by what actually weighs. A structured document with overview, data, metrics, and risks. When it finishes, the agent also offers **strategic angles** — new questions raised by the investigation itself — and the option to **turn it into an action plan**. A discovery can take from a few minutes to several dozen, depending on depth. It runs on the server: you can close the screen and come back later.