> ## Documentation Index
> Fetch the complete documentation index at: https://docs.simulant.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# Research agent

> Ask questions of your simulation in plain language and get grounded answers back.

The **research agent** lets you interrogate your simulation in plain language. Instead of building a study for every question, you can ask directly — and the agent draws on your workspace's cohorts, tests, and grounding data to answer.

## Open the agent

Select **Agent** at the top of the sidebar. Type a question and the reply streams back as it's generated.

<Frame caption="The research agent — plain-language answers with the cohorts, tests, and verbatims each response draws on.">
  <img src="https://mintcdn.com/simulant/j7X70BC9NUZuF_VF/images/app/agent.png?fit=max&auto=format&n=j7X70BC9NUZuF_VF&q=85&s=110f31e6e84f6fc44f92691c2fc7940f" alt="Research agent answering which voter segment reacted worst to a cost-of-living spot, with source chips" width="1428" height="779" data-path="images/app/agent.png" />
</Frame>

## What to ask

The agent is most useful for questions that sit across your work rather than inside a single test:

* "Which segment reacted most negatively to the Cost-of-Living spot, and why?"
* "Summarize the recurring themes across our last three message tests."
* "What would a cost-of-living-anxious cohort likely think about this headline?"
* "Where do our suburban swing and rural independent audiences disagree?"

## Grounded answers

The agent answers from the same grounded record the rest of Simulant is built on, so responses trace back to real signal rather than guesswork. Treat it as a way to explore and narrow a question quickly — then run a [test](/guides/tests) or a [focus group](/guides/focus-groups) when you need a rigorous, shareable result.

<Tip>
  Ask follow-ups. The agent keeps the thread, so you can drill from a broad question down to a specific segment without restating context each time.
</Tip>