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

# Reports Agent

> Ask a question about your conversations and get the answer, the evidence, and the file — by chatting with Metaview's agent.

You already have the question in mind: Turning that question into a report means knowing which filters narrow the conversations, which columns surface the answer, and when a question needs an AI column to pull it out of a transcript. The Reports Agent lets you skip the setup work and get the answer back in a quick conversation.

The Reports Agent is a chat panel that lives on every report, as well as on the New Report page. Ask a question about your conversations and it builds or updates the report, reads the transcripts when it needs to, and comes back with the number, the breakdown, the links to the conversations behind the answer, and any file you asked for.

<Frame caption="The Reports Agent opens in the Assistant tab of any report, and on the New Report page." />

## How the Reports Agent Works

Open any report and go to the `Assistant` tab, or start a new report from Reports → `New report` and type your question into the prompt box.

<Steps>
  <Step title="Describe what you want to know">
    Type a plain-language question — *"What are candidates' main hesitations about joining?"*, *"Average interviewer talk time on job interviews in the last 90 days, by interviewer"*, *"Which recruiter screens mentioned our competitor?"* You can also use voice input, or attach context like a job description, interviewing framework, or rubric to shape the answer.
  </Step>

  <Step title="Let the agent build the report">
    The agent updates the report in place: filters, grouping, metrics, columns, and new AI columns if the question needs them. On questions that need a new AI column, it runs the column on a sample of 20 conversations first and asks before running it over your full set — so you can sanity-check the column before it works through the rest.
  </Step>

  <Step title="Read the answer">
    The agent answers in chat with the number *and* its denominator — *"595 of 1,529 interviews (39%) raised a hesitation, role scope the biggest theme"* — plus a breakdown and links to the conversations behind it. The underlying report shows the same basis, so you can click into any row to see the source.
  </Step>

  <Step title="Get the deliverable you need">
    Ask for a CSV, PDF, chart, or Metaview-styled dashboard, and the agent produces it as a card in the chat. `"Show me that call"` opens a card with the specific conversation. You can also ask the agent to pull in outside context from the public web — a company's funding round, a market stat — when a question needs it.
  </Step>

  <Step title="Spin off a follow-up report">
    When a follow-up needs a different population than the current report covers — a wider date range, a different conversation type — the agent proposes a separate new report as a draft. You create it with one click, and the new report opens in its own chat with a handoff note explaining what it's for.
  </Step>
</Steps>

Everything behind these steps — how filters, grouping, columns, charts, sharing, subscribers, and CSV download work — is covered in [Reports Overview](/reports/overview).

<Info>
  As you drill into a grouped row, the drill-down view shows a narrower slice of the same report, and the agent chat stays with you — it doesn't fork into a separate conversation per drill-down.
</Info>

## AI columns are workspace-wide

When the agent creates an AI column to answer your question, that column becomes part of your whole workspace's column library — not just this report. Three things worth knowing:

* Anyone adding columns to any report in the workspace will see it in the `Add column` menu.
* Other reports that already use the column will show the agent's version of its results.
* If you later ask the agent to change the column's definition, the change clears the column's results in every report that uses it, and the agent will ask you first before making that change.

If you want a one-off calculation that doesn't persist across the workspace, say so — the agent can filter and compute the answer without creating a new column where possible.

## Who can see and send to the chat

The agent chat is shared across everyone the report is shared with, so you can see each other's conversations and the agent's answers:

* **Everyone who can see the report can read the chat.**
* **Only people who can edit the report can send messages to the agent.** If you're viewing a report in read-only mode, you'll see the chat history but not a message box.
* **"Created by Metaview" default reports are read-only for the agent chat** — you can still ask Metaview questions by duplicating the report into one you own.
* The agent only ever sees what you can see. It respects the report's existing access rules; it doesn't pull in conversations the viewer wouldn't otherwise see.

## What the agent won't do

The Reports Agent reports what the data shows. It doesn't make hiring recommendations, rank or judge candidates as a fit for a role, or use or infer protected characteristics (such as race, gender, age, or nationality) in its answers. If you ask it to do one of these, it will tell you it won't and offer a factual alternative.

***

Hit a problem, or looking for a more specific answer? See the [Reports Overview](/reports/overview) for how filters, columns, charts, sharing, and subscribers work.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.