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

# Agent

> What Kadoa's agent does, its two chats, and when it needs you

Kadoa's agent builds and maintains your workflows. You describe the data you need. The agent finds the source, writes the extraction, tests it, and keeps it running when sites change.

## Two chats

Select **Agent** on any page to open it. The panel has two chats, and switching between them changes what the agent works on.

![The agent panel side by side: a private Workspace chat with a source research request, and the This workflow chat that everyone on the team can see](https://mintcdn.com/kadoa/v7EEuXHZHS8J6dQ7/images/assistant/agent-scopes.png?fit=max\&auto=format\&n=v7EEuXHZHS8J6dQ7\&q=85\&s=aa5341cdbdd3955efe2766ae139c73fc)

| | Workspace | This workflow |
| - | - | - |
| Who sees it | Only you | Everyone on your team |
| What it covers | All of your team's workflows | One workflow |
| Use it to | Research sources, create workflows (also many at once), and ask questions across workflows | Change, fix, or debug this workflow, and answer the agent's questions about it |
| History | Your own chats, under **History** | One shared chat per workflow |

On a workflow page, the panel opens on **This workflow**. A dot on **Agent** means a workflow is waiting for your reply.

## What the agent does

* Finds the source, if you don't give a URL.
* Proposes a schema and picks [key fields](https://docs.kadoa.com/docs/workflows/key-fields.md), or uses the fields you name.
* Writes the extraction code and tests it on a free preview.
* Repairs the workflow when a run breaks or a site changes.

You can also reach the agent from code or from your AI tool over [MCP](https://docs.kadoa.com/docs/sdk/mcp.md). Build and repair requests go to the same agent.

## How runs work

The agent writes each workflow's extraction code once. Every run executes that code, so runs are repeatable and do not depend on a new AI guess each time. The agent steps in again only when you ask for a change or when a run needs a repair.

## When the agent needs you

* **Reply needed**: the agent has a question. Answer it in the workflow's chat.
* **Preview ready**: check the sample and approve it. See [Preview and approval](https://docs.kadoa.com/docs/ui/workflows/human-in-the-loop.md).

If the agent cannot fix a problem, it hands the problem to [Kadoa support](https://docs.kadoa.com/docs/support-center.md).

## Learn more

* [Work with the agent](https://docs.kadoa.com/docs/workflows/agent.md): step-by-step tasks in the dashboard, from code, and over MCP.
* [AI personalization](https://docs.kadoa.com/docs/ui/ai-personalization.md): rules the agent follows on every new workflow.


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