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

# Agent Configuration

> There are different tools you can use to configure the agent and tailor it to your specific use case.

<Tip>
  For more details you can read our detailed [agent creation guide](/en/using-langdock/guides/agents/agent-creation) of how to build an agent.
</Tip>

Instead of setting the options below manually, you can describe your agent in the [Agent Builder chat](/en/using-langdock/agents/agent-builder) and let it configure them for you. Both approaches work on the same configuration, so you can switch between them at any time.

You have the following configuration options to customize your agent:

### Icon, Name, and Description

Short descriptive information to identify the agent and describe how it works to other users. Workspace admins can also set a [creator display name](/en/using-langdock/agents/advanced-features#creator-display-name) that is shown instead of the individual creator, for example a team name.

| Property           | Limit             |
| ------------------ | ----------------- |
| Agent name         | 80 characters     |
| Agent description  | 500 characters    |
| Agent instructions | 40,000 characters |

### Input Type

Agents support different input types that determine how they receive and process information.

#### Prompt (Default)

The chat input field you already know from normal chat. This lets you send any message to the agent and receive a response.

You can set conversation starters, which are saved prompts users can click instead of writing the first message. These help guide users and reduce the effort needed to get started.

#### Form

Forms collect information in a structured way, similar to a survey tool. This helps guide users to understand how much context is necessary for a quality response and collects it in a standardized format that the model can process more easily.

### Instructions

Describe what you want to achieve with this agent and define clear instructions. Include as many relevant details and background information as possible. This enables the agent to answer better and more closely to your expectations. Check out our [agent creation guide](/en/using-langdock/guides/agents/agent-creation) and our [prompting guide](/en/using-langdock/guides/prompt-engineering/basics/prompt-elements) for more details.

### Knowledge

Attach knowledge and files to your agent. You can either upload files from your computer or attach files from the [integrations](/en/using-langdock/guides/integrations/using-integrations) to the agent. Attached Folders give the Agent context that it cannot edit.

#### Source Access Restriction

By default, when your agent cites sources in its responses, users can click on those references to open and view the original documents. If you need to protect sensitive source content while still allowing the agent to reference it, you can enable source access restriction.

When enabled, users see source references in responses but cannot click to open them. This is useful when:

* The agent needs access to confidential documents to provide accurate answers
* Users should receive information from sources without direct access to the underlying files
* You want to control document distribution while still leveraging the knowledge

<Info>
  This setting must first be enabled at the workspace level by an admin before it becomes available in the agent configuration. If you don't see this option, contact your workspace admin.
</Info>

### Actions

Actions extend what your agent can do. Click **Add action** to add any of the following:

#### Capabilities

Built-in features from the chat:

* **Web Search** - search the internet for current information
* **Image Generation** - create images from text descriptions

Agents can create files, edit file content, and analyze tabular data automatically when this is enabled in the workspace settings.

#### Integration Actions

Connect your agent to external tools and services. Your agent can then perform actions like:

* Create email drafts
* Update CRM entries
* Create support tickets
* Post messages to Slack
* And many more - see our [integrations guides](/en/using-langdock/guides/integrations/using-integrations)

#### Folders

Attach [Folders](/en/using-langdock/library/folders) when you want the Agent to use files from a folder as context. Attached Folders give the Agent context, but the Agent cannot edit the folder or save files back to it.

#### File Templates

Attach [file templates](/en/using-langdock/library/file-templates) so your agent generates documents and presentations in a consistent format and style.

#### Workflows

Attach [workflows](/en/using-langdock/workflows/introduction) so your agent can trigger multi-step automations during a conversation. See [using workflows through agents](/en/using-langdock/chat/tools/workflows#using-workflows-through-agents) for details.

#### Other Agents

Attach other agents to enable delegation. Your agent can call specialized agents to handle specific subtasks, allowing you to build complex multi-agent workflows.

<Warning>
  Deep Research is only available in regular chats, not when using agents. To use [Deep Research](/en/using-langdock/chat/tools/deep-research), switch to a regular chat session.
</Warning>

### Model

Choose which model this agent will use. For details about choosing the right model, refer to our [model guide](/en/using-langdock/models-and-limits/models).

### Creativity

Controls the temperature parameter of the model, which affects how deterministic or creative responses are. The slider ranges from 0 (deterministic) to 1 (creative):

* **Lower values (0-0.3)**: More focused, consistent, and predictable responses. Best for factual tasks, coding, or when you need reliable outputs.
* **Medium values (0.4-0.7)**: Balanced creativity and consistency. Good for general use cases. Default is 0.7.
* **Higher values (0.8-1.0)**: More varied and creative responses. Better for brainstorming, creative writing, or when you want diverse outputs.

### When Limits Are Reached

Choose what happens when a [user](/en/admin/manage-usage/byok/usage-limits#per-user-limits), [workspace](/en/admin/manage-usage/byok/usage-limits#workspace-spend-limit), or [agent limit](/en/admin/manage-agents/agent-limits) is reached. **Switch model** (default) lets the agent answer with the fallback model; **Block usage** makes the agent unavailable until the limit resets.

### Schedule

Use the **Schedule** button in the top bar to create a [scheduled task](/en/using-langdock/chat/scheduled) with the agent preselected, for example a daily briefing that runs through this agent.

### Sharing

In the top right corner, you'll find options to share and use the agent. You can share it with anyone in the workspace or assign editing and usage permissions to specific groups or individuals. If the agent has [subagents](/en/using-langdock/agents/subagents) attached, Langdock checks whether the people you share it with can also access those subagents and offers to [grant missing access](/en/using-langdock/agents/subagents#permissions-and-access).

### Publishing

Changes to an agent's configuration are saved as a **draft** and do not affect users until you publish. Click **Publish** in the top right corner to make the current draft the active version. The Agent API endpoints always use the active version of an agent.

When publishing a new version, you can add an optional update message describing what changed. Users then see a **New version** notice when they start a new chat with the agent. The notice disappears 7 days after publishing, or earlier once a user has started 3 chats with the agent since the release. You don't see the notice for versions you published yourself, and the first published version never shows one.

### Analytics

Use the **Analytics** and **Feedback** tabs to review agent performance and user feedback. See [Agent Analytics](/en/using-langdock/agents/analytics) for the full Analytics and Feedback reference.

### Tracing and Logging

For deeper insights into your agent's behavior and performance, you can enable tracing through [Langfuse](https://langfuse.com).

**How it works:**

1. **Workspace admins** first enable agent logging in the workspace settings. This makes the feature available for agents in your workspace.

2. **Agent editors** can then configure logging for individual agents:
   * Enable the **Allow agent logs** toggle in the agent configuration
   * Set the **Tracing cloud URL** (defaults to `https://cloud.langfuse.com`)

Once configured, detailed logs about your agent's interactions are sent to Langfuse, giving you visibility into:

* Individual conversation traces
* Model inputs and outputs
* Performance metrics and latency
* Token usage per interaction

<Info>
  This feature requires workspace admin approval before it can be used. If you don't see the logging options, contact your workspace admin to enable agent logs in the workspace settings.
</Info>

<Tip>
  Langfuse offers both a cloud version and self-hosted options. If your organization uses a self-hosted Langfuse instance, update the Tracing cloud URL to point to your internal deployment.
</Tip>

<Info>
  For additional agent management features like labels, pinning, duplication, and owner transfer, see [Advanced Features](/en/using-langdock/agents/advanced-features).
</Info>

## FAQ

<AccordionGroup>
  <Accordion title="What information counts toward an agent's context?">
    An agent's context can include the user's message, previous conversation, agent instructions, attached files, retrieved knowledge, available tools, tool results, and system instructions. Complex agents can therefore fill context faster than a simple chat.
  </Accordion>

  <Accordion title="How should I decide which knowledge sources and actions to attach to an agent?">
    Attach only the knowledge and actions the agent needs for its specific job. Broad agents with many sources and tools are harder to control and can retrieve irrelevant information. For complex work, split responsibilities across focused agents or subagents.
  </Accordion>

  <Accordion title="Why does an agent not use an attached tool or knowledge source?">
    An agent may skip a tool or source if the prompt does not make the need clear, if permissions are missing, if retrieval does not find relevant content, or if another instruction has higher priority. Make the expected workflow explicit in the agent instructions and test with representative examples.
  </Accordion>

  <Accordion title="How does attached knowledge affect an agent's answers?">
    Attached knowledge gives the agent reusable information to retrieve during a conversation. The agent usually receives relevant excerpts, not necessarily every attached document in full. The quality of the answer depends on retrieval, permissions, instructions, and the user's question.
  </Accordion>

  <Accordion title="When should I use an agent-specific Knowledge base instead of direct files?">
    Use an agent-specific Knowledge base when the same documents should support repeated conversations or many users. Use direct files when the information is temporary, small enough for the current task, or only needed in one chat.
  </Accordion>

  <Accordion title="Why does an agent respond in the chat instead of creating a document?">
    If an agent's instructions require responding directly in the chat or forbid creating files, the Document Editor won't open automatically. An explicit request for a document in the message still creates one, and users can always open the editor from the **Create document** button or the **Tools** menu.
  </Accordion>
</AccordionGroup>