---
title: "Using the Agent"
canonical: https://trickest.com/docs/using-the-app/agent/using-the-agent
description: "Open the Agent, start a conversation, choose a model, connect a workflow, and respond when it needs you."
---

# Using the Agent

## Before you start

The Agent needs access to a model, either through [Managed AI](/docs/key-concepts/agent/managed-ai) or a verified [provider key](/docs/using-the-app/agent/bring-your-own-key). If **Agent** is missing from the sidebar, first confirm that the workspace has model access.

## Open the Agent

Select **Agent** in the sidebar. That opens the Agent workspace.

In a workflow's editor, select **Agent** in the bottom chat dock to open a conversation. **Recent chats** in that dock reopens an existing conversation.

## Start a conversation

Write what you want done, then select **Send** or press **Enter**. **Shift+Enter** adds another line.

To continue an earlier conversation, open the title next to **Agent** and choose it. On a new chat, recent conversations also appear under the composer. The switcher shows **Running**, **Needs input**, or **Queued** for active conversations; open a chat marked **Needs input** to answer its question or permission request.

## Choose a model

The model control is in the composer, next to the workflow control. Open it and select a model your workspace can use.

If no model is listed, the workspace still needs [Managed AI](/docs/key-concepts/agent/managed-ai) or a verified [provider key](/docs/using-the-app/agent/bring-your-own-key).

## Connect a workflow

Connect a workflow when the Agent should work on a specific graph. The composer control is labeled **Workflow** until one is attached. After you choose one, that workflow opens beside the conversation so you can watch the work.

You can also start without a workflow and connect one later. Connecting a workflow changes the graph context; it does not move files or automatically switch the Agent's device.

## Choose where the Agent works

Open **Select device** in the workspace to inspect an available device. Viewing one does not assign it to the Agent: use **Use this device for agent** to change where subsequent work runs, or **Follow agent** to return to its current device. Save modified files before switching. Commands already running remain on their original device. See [Managing Devices](/docs/using-the-app/computers/managing-computers).

## Work with the Agent

Follow the reply as the Agent works, then review the workflow or files beside the chat. Treat the reply as a report of the work and use the resulting graph or files to confirm the result.

While a reply is in progress, you can type a follow-up. **Enter** sends that note for application at a safe point in the current turn. Its receipt distinguishes waiting from added to context; delivery does not mean the requested work has finished. **Stop** ends the current turn, but it may not stop work the Agent has already started elsewhere. A workflow run or long-running command can continue, so check its state before sending the same request again.

If the Agent asks a question, answer it in the thread. If it needs permission, the card says **Permission required**. Choose **Allow** or **Deny**.

If it proposes a plan, read the summary first. **Start** tells it to carry out the plan. **Cancel** declines it. You can type a change in the composer before you start.

For an objective that should continue across turns, [create and manage a goal](/docs/using-the-app/agent/working-with-goals). To consult or continue work in another of your chats, use [conversation coordination](/docs/using-the-app/agent/coordinating-conversations).

## Continue from here

- [Working with Files, Terminal, and Previews](/docs/using-the-app/agent/files-terminal-previews) for the files and shell beside the chat.
- [Using the Device Screen](/docs/using-the-app/agent/using-the-computer) for the sandbox desktop.
- [Creating and Importing a Skill](/docs/using-the-app/agent/creating-a-skill) for reusable instructions.
- [Scheduling an Automation](/docs/using-the-app/agent/scheduling-an-automation) for a prompt that runs on a cadence.
- [Adding an AI Agent Node](/docs/using-the-app/agent/adding-an-ai-agent-node) for an agent step on the workflow graph, which is separate from this chat.

For a scoped external-data request and a reviewed workflow proposal, follow [Use MCP connectors with the Agent](/docs/using-the-app/agent/using-mcp-connectors).

---
_Markdown source of https://trickest.com/docs/using-the-app/agent/using-the-agent._