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

# Build with NexorGPT

NexorGPT is the chat inside Nexor. Think of it as a consultant that asks the questions needed to configure your agent well, then applies the answers for you. Anything you can change in the visual settings, you can also change here.

<Frame caption="The NexorGPT page, with one-tap suggestions when the chat is empty.">
  <img src="https://mintcdn.com/nexor/335V9P_gdO6wPMqv/images/getting-started/nexorgpt-page.png?fit=max&auto=format&n=335V9P_gdO6wPMqv&q=85&s=4968dec0f48f9c1fc93d349450f69ff9" alt="NexorGPT page with quick suggestions" width="1600" height="900" data-path="images/getting-started/nexorgpt-page.png" />
</Frame>

## Where to find it

* **The NexorGPT page** in the sidebar, for longer work like building an agent.
* **The panel on the right** of most screens, for quick edits and questions while you work. It is not available in the playground or in the manual create-agent wizard, which have their own tools.

Only you talk to NexorGPT. It never messages your leads.

## Create an agent by chat

<Steps>
  <Step title="Start from Create agent">
    Click **Create agent** and choose **NexorGPT** instead of the manual flow. Instagram and Messenger automations keep the manual flow.
  </Step>

  <Step title="Answer the questions">
    NexorGPT asks the same questions the Nexor team asks in onboarding: what to achieve, where leads come from, which channels, what to collect, how to qualify, and what counts as done. Paste answers from your [worksheet](/docs/en/guides/getting-started/define-your-use-case). Resolve open questions on the spot rather than leaving them for later.
  </Step>

  <Step title="Review and confirm">
    NexorGPT summarizes the agent it will create. Read it as if it were a job description for a new hire and correct anything that is off.
  </Step>

  <Step title="Keep the conversation open">
    After the agent is created, the same chat stays open. Use it for the adjustments you will want after your first test.
  </Step>
</Steps>

<Frame caption="NexorGPT collects the brief in chat and creates the agent from it.">
  <img src="https://mintcdn.com/nexor/335V9P_gdO6wPMqv/images/getting-started/nexorgpt-create-agent.png?fit=max&auto=format&n=335V9P_gdO6wPMqv&q=85&s=f3ed2021578ffea848289f0f62e66560" alt="NexorGPT asking questions to create an agent" width="1600" height="900" data-path="images/getting-started/nexorgpt-create-agent.png" />
</Frame>

<Tip>
  Describe the outcome and the constraints, not the exact words the agent should say. NexorGPT and the prompt sections handle the phrasing; you provide the business.
</Tip>

## Keep it structured, not one giant prompt

A single long prompt that tries to cover every situation gets harder to test and easier to break with every edit. Two habits prevent that:

* **Use the sections.** Business context, limits, objections, escalation, examples, additional rules, and discard rules each have a place. Put each instruction where it belongs.
* **Split by intention.** If the process has different flows, such as qualifying new leads and then sending a payment link, build one agent per flow and connect them with transfers. See [Connected agents](/docs/en/guides/getting-started/connected-agents).

## Edit an agent by chat

Once the agent exists, describe the change in plain language. NexorGPT shows a before and after card, and nothing changes until you confirm. Prompt changes can be undone from the same card.

<Frame caption="Every edit shows a before/after card before it is applied.">
  <img src="https://mintcdn.com/nexor/335V9P_gdO6wPMqv/images/getting-started/nexorgpt-edit-confirmation.png?fit=max&auto=format&n=335V9P_gdO6wPMqv&q=85&s=3b73a88bc71f91ff8ab182ab59a4d00f" alt="NexorGPT before/after confirmation card" width="1600" height="900" data-path="images/getting-started/nexorgpt-edit-confirmation.png" />
</Frame>

Examples of things to ask:

* "Add a required variable for budget range, as a select with three options."
* "The agent should never quote final prices. Refer pricing questions to a person."
* "Move leads to Lost after three unanswered follow-ups instead of five."
* "Change the first-contact channel to email."

<Note>
  When several agents exist, NexorGPT edits the one you are looking at or the one you name. Check which agent the card refers to before confirming. See [Edit your agent by chat](/docs/en/guides/assistant/editing-your-agent).
</Note>

## Ask about what the agent knows and how it is doing

NexorGPT can read the agent's setup and your data on request. Useful questions while onboarding:

* "What does this agent know?" to check the knowledge base.
* "What is in my catalog?" when you sell from one.
* "What can I improve in my agent?" for suggestions after a few test conversations.
* "Why did the agent answer this way?" with a lead attached, once real conversations exist. See [Debug a lead](/docs/en/guides/assistant/debugging-leads).

## Next

Your agent exists. Go to [Test in the playground](/docs/en/guides/getting-started/test-in-playground) before showing it to any lead.