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

# Agents

> Configure AI teammates that execute real multi-step work across apps and websites

## Overview

**Agents** are AI teammates capable of executing real multi-step work across apps and websites.

Agents can read and use your organization's knowledge, follow rules you define, and work inside workflows to complete complex tasks end-to-end. You can configure what they know, how they should behave, and then chat with them or embed them inside **Workflows** to automate processes.

<CardGroup cols={2}>
  <Card title="Organizational Knowledge" icon="books">
    Attach **Library Files** so the Agent answers from your own indexed documents
  </Card>

  <Card title="Rules and Guardrails" icon="shield-check">
    Define the rules the Agent must follow and keep answers grounded in the selected context
  </Card>

  <Card title="Web Enrichment" icon="globe">
    Optionally turn on **Web Search** when the Agent needs external information
  </Card>

  <Card title="Workflow Ready" icon="diagram-project">
    Chat with an Agent directly, or embed it inside a **Workflow** to automate a process
  </Card>
</CardGroup>

## Creating an Agent

<Steps>
  <Step title="Create the Agent">
    Click **Create Agent** under **Agents**.
  </Step>

  <Step title="Name and Describe It">
    Give your Agent a clear **Name** and **Description** so users can easily recognize its role and purpose.
  </Step>

  <Step title="Assign a Department">
    Assign a **Department** label to specify which department this Agent belongs to. Use consistent labels across Agents so teams can easily filter and find the right one.
  </Step>

  <Step title="Attach Context from Library Files">
    Attach context from **Library Files**. Select one or more relevant files ingested in the **Library**. Ensure only high-quality, up-to-date documents are selected to avoid outdated or conflicting answers.
  </Step>

  <Step title="Enable Guardrails">
    Tick the **Guardrails** checkmark to enable information retrieval from the **Selected Library Files**.
  </Step>

  <Step title="Define the Rules">
    Define the rules the Agent must follow, for example: *Answer only from the selected Library files. If information is not in the context, say you don't know.*
  </Step>

  <Step title="Turn On Web Search (optional)">
    Turn on **Web Search** if you want the Agent to enrich answers with information from the web.

    <Warning>
      If you do not enable **Guardrails**, the Agent may use web sources and could respond beyond the specific documents or rules you intend. Always enable **Guardrails** for controlled, compliant behavior.
    </Warning>
  </Step>

  <Step title="Instruct Your Users">
    Specify the **Description** to instruct users what input to specify when chatting with the Agent.
  </Step>

  <Step title="Test and Refine">
    Start a conversation by asking it questions that match its purpose. Verify that the Agent uses the selected **Library files** accurately while following the **Guardrail** rules. Then refine them as needed based on its responses.
  </Step>
</Steps>

## Agents in Workflows

Agents are not limited to chat. You can embed them as a step inside a workflow, or chain several of them together:

* Embed **Existing Agents** in the workflow to perform the pre-defined instruction and rules bound by the Agent.

* Multiple **Existing Agents** can be linked together in a chain using the Workflows. Linking one Agent to another allows you to design "teams" of AI teammates, where each Agent handles a specialized part of the process. Configure inputs/outputs so the next Agent can use results produced by the previous Agent. For example, Agent A summarizes a document, Agent B reviews the summary for compliance.

<Info>
  To run an agentic task that reasons, executes code, and generates files inside a workflow step, use the [Agent action node](/how-to-use/workflows#agent).
</Info>

## Pro Tips

<Tip>
  **Treat Agents as AI teammates that you are managing.** Give them a clear role: "HR assistant", "Sales deck generator", "Policy Q\&A".
</Tip>

<Tip>
  **Guardrails first, Web Search second.** Start with **Guardrails** and **Selected Library Files** only to ensure accuracy. Turn on **Web Search** only when you explicitly need external information (e.g., market research, news).
</Tip>

<Tip>
  **Curate context carefully.** Avoid adding too many unrelated **Library Files** to a single Agent. Group context by topic or use case to reduce conflicting or noisy answers.
</Tip>

<Tip>
  **Show users how to prompt the Agent.** When explaining how to use the Agent in the Description, include examples such as: "Ask me: 'Summarize the Q4 policy changes for sales reps.'" "Use me when you need: draft responses based on the Employee Handbook."
</Tip>

<Tip>
  **Test before sharing** by chatting with the Agent yourself using realistic and irrelevant questions. Check for accurate citation and if it behaves correctly when it doesn't know the answer (e.g., admits uncertainty instead of guessing).
</Tip>

## Next Steps

<CardGroup cols={3}>
  <Card title="Knowledge Base" icon="books" href="/how-to-use/library">
    Ingest and index the files your Agents will use as context
  </Card>

  <Card title="Workflows" icon="diagram-project" href="/how-to-use/workflows">
    Embed and chain Agents inside automated processes
  </Card>

  <Card title="Chatbots" icon="robot" href="/how-to-use/chatbots">
    Publish a conversational experience for external users
  </Card>
</CardGroup>