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

# Chat with a custom agent

> Open a live conversation with one of your custom agents — it answers and works under its own instructions, scope, and execution mode.

Custom agents usually work on their own — a schedule fires, the agent runs, a report lands. You can also talk to one directly. A chat with a custom agent is a live conversation with the specialist you briefed: it follows the agent's instructions, focuses on the agent's data scope, and respects what the agent is allowed to do.

## When to use this

* **Ask an agent about itself.** "What do you watch?", "When do you run?", "What are you allowed to change?" — the agent answers from its live configuration, not a guess.
* **Dig into the agent's territory.** Investigate a late delivery or a provider trend with the agent's playbook and scope already applied, instead of re-briefing the main assistant from scratch.
* **Follow up on a run.** A scheduled report flagged something; ask the specialist behind it what to make of the finding.
* **Try out a new brief.** Before you put an agent on a schedule, chat with it to see how it interprets its instructions.

For a one-off question outside any agent's territory, just talk to the main Nash assistant — a scoped agent focuses on its own slice of your operation.

## Start a chat

There are two ways in, and both bind the conversation to the agent from the first message:

* **From the agent's page.** Open the agent under **Agents** and select **New chat**. The new conversation is labeled with the agent's name so you always know who you're talking to.
* **Mention it in the message box.** Type `@` where you'd normally write to Nash, pick the agent, and send. That message starts a new conversation bound to the agent.

A conversation keeps the binding it started with. Mentioning an agent in the middle of an existing thread doesn't hand the conversation over — start a new chat to switch specialists.

## What carries over from the agent

Everything you configured on the agent shapes the conversation:

| From the agent                   | In the chat                                                        |
| -------------------------------- | ------------------------------------------------------------------ |
| **Instructions**                 | The agent approaches your questions with its brief in mind.        |
| **Scope**                        | It focuses on the providers, zones, or contracts you scoped it to. |
| **Skills, tools, and knowledge** | The same capabilities and reference material its runs use.         |
| **Execution mode**               | Whether it can make changes from the chat — see below.             |

The agent can also describe its own setup accurately — its purpose, scope, schedules and triggers, and what it's permitted to do — so "what would you do if a delivery goes late?" gets an answer grounded in how the agent is actually configured.

## Whether the agent can act

The agent's [execution mode](/nash-agent/custom-agents#set-it-up) carries into the conversation:

* **Report only** — the chat is advisory. The agent investigates, answers, and tells you what it *would* do, but it can't make changes from the conversation — no matter how the request is phrased. If you want the agent acting, update its execution mode and start a new chat, or ask the main Nash assistant to make the change.
* **Auto-execute** — the agent can carry out actions during the conversation, and sensitive actions pause for your approval first, exactly as in any Nash Agent chat. See [Guardrails & confirmations](/nash-agent/guardrails).

<Note>
  The mode is a property of the agent, enforced on every turn — you can't loosen it from inside a conversation.
</Note>

## FAQ

<AccordionGroup>
  <Accordion title="Does chatting with an agent count as one of its runs?">
    No. A chat is a conversation you own, saved in your chat history like any other. Scheduled and triggered runs are separate — talking to an agent doesn't consume a run or fire its schedule.
  </Accordion>

  <Accordion title="Who can chat with a custom agent?">
    Anyone who can chat with Nash. You don't need permission to manage agents — managing and chatting are separate; see [who can create and manage custom agents](/nash-agent/custom-agents#faq).
  </Accordion>

  <Accordion title="Can I change the agent's configuration from the chat?">
    The conversation follows the agent's saved configuration. To change what the agent does — its instructions, scope, or execution mode — edit the agent itself.
  </Accordion>
</AccordionGroup>

## Related

<CardGroup cols={2}>
  <Card title="Custom agents" icon="user-gear" href="/nash-agent/custom-agents">
    What a custom agent is, and how to set one up for a recurring job.
  </Card>

  <Card title="Guardrails & confirmations" icon="shield-check" href="/nash-agent/guardrails">
    How Nash decides which actions run automatically and which wait for your approval.
  </Card>

  <Card title="Scheduling and execution" icon="clock" href="/nash-agent/scheduling-and-execution">
    Run an agent on a schedule, on an alert, or on a file upload.
  </Card>

  <Card title="Knowledge" icon="book" href="/nash-agent/knowledge">
    Ground your agents with SOPs, policies, and reference links.
  </Card>
</CardGroup>