# Agent node

> Configure autonomous planning, knowledge retrieval, tools, and Skills inside a workflow.

Source: https://doc.fastgpt.io/en/guide/build/workflow/nodes/agent

The Agent node (`agent`) embeds a ReAct-based autonomous planning loop inside a deterministic workflow canvas, letting your workflow maintain strict outer routing while delegating an open-ended sub-task to an Agent equipped with tools, knowledge bases, a virtual machine sandbox, and Skills.

The **Basic** tab of the current node panel does not list a standalone Agent node. The **Agent** tab lists existing applications, which are added to the canvas as [Call Application](/en/guide/build/workflow/nodes/run_app) nodes. To try autonomous tasks and Skills, start with [your first Agent task](/en/guide/getting-started/first-agent) in Chat Agent V2.

## Inputs and configuration

| Setting            | Internal Key      | Type            | Purpose                                                                          |
| ------------------ | ----------------- | --------------- | -------------------------------------------------------------------------------- |
| **AI Model**       | `model`           | `String`        | Select a tool-calling model                                                      |
| **Prompt**         | `systemPrompt`    | `String`        | Define task goals, output format, and tool-use boundaries                        |
| **User Question**  | `userChatInput`   | `String`        | Reference the sub-task instruction from the start node or an upstream node       |
| **File Link**      | `fileUrlList`     | `Array<string>` | Supply document or image URLs accessible to the current task                     |
| **Select Dataset** | `datasets`        | `SelectDataset` | Bind datasets and search parameters for on-demand Agent retrieval                |
| **Tools**          | `selectedTools`   | Tool list       | Attach system tools, HTTP tools, MCP tools, or workflow tools                    |
| **Skill**          | `skills`          | Skill list      | Associate published [Agent Skills](/en/guide/build/skill/integration)            |
| **Computer**       | `useAgentSandbox` | `Boolean`       | When enabled, run scripts and file operations inside the virtual machine sandbox |

## Outputs and connections

| Output Port             | Internal Key        | Value Type | Description                                                                                                       |
| ----------------------- | ------------------- | ---------- | ----------------------------------------------------------------------------------------------------------------- |
| **AI Response Content** | `answerText`        | `String`   | Final conclusion text synthesized after multi-step reasoning and tool calls                                       |
| **Error text**          | `system_error_text` | `String`   | Available when **Error catch** is enabled; outputs diagnostic error details when model or sandbox execution fails |

Execution connections control run order; value references pass `answerText` to downstream Condition, Text Editor, or Assigned Reply nodes.

![Agent and tool connection example](https://cdn.jsdelivr.net/gh/labring/fastgpt-img@main/optimized/current/tool-connections.png.webp)

## Example: policy checklist (legacy workflows)

The steps below apply to legacy workflows that already contain a standalone **Agent node**. For new workflows, create a Chat Agent V2 following [your first Agent task](/en/guide/getting-started/first-agent) and add it to the canvas from the **Agent** tab as a [Call Application](/en/guide/build/workflow/nodes/run_app) node.

1. Connect **Workflow Start** to the **Agent node** and reference `userChatInput` as its **User Question**.
2. Associate the `Travel Policy` knowledge base and instruct the Agent: `Retrieve the travel policy first, then list submission deadlines, required receipts, and finance review timelines.`
3. Reference `AI Response Content` in a downstream archiving or Assigned Reply node.
4. Debug with `Prepare a travel expense checklist` and expand the Agent node in run details to verify its retrieval call, returned documents, and deadlines. See [your first Agent task](/en/guide/getting-started/first-agent) for sample material.

Put steps with a required order in workflow connections. Tool selection inside the node depends on model planning. External write operations need clear permissions and confirmation requirements. Investigate failures through execution details, model tool support, resource permissions, and virtual-machine status.

Next: [Skill integration](/en/guide/build/skill/integration) and [published versions](/en/guide/build/publish/versions).