Agent Builder - Advanced Configuration¶

The module system and every optional module you can add to an agent.

Overview¶

Beyond the mandatory Identity and Brain core, everything in the Agent Builder is a module — a self-contained section you add only when you need it.

This guide covers the module system itself and then each module in turn. For the core setup, see Agent Builder - Basic Configuration.

The Module System¶

Adding a Module¶

Click Add a section in the left Agent sections sidebar. Search by section name or browse the groups, then click Add beside a section. Added sections remain listed with an Added indicator.

Modules are grouped by intent — Capabilities, Knowledge, Distribution and Tuning

Modules are grouped by intent:

Widget and Floating Widget Auth are unavailable until Floating Widget is enabled in Channels.

The Agent Sections Sidebar¶

A sidebar on the left lists every section currently on the agent and navigates to it.

• Identity and Brain carry a lock icon — they are mandatory and cannot be removed
• Every other module carries a drag handle
• Drag the handle to reorder sections, or focus it and press the arrow keys

Removing a Module¶

Open an optional section's actions menu to remove it. Removing it takes its configuration off the page.

Brain — Advanced Instructions¶

Brain holds more than the main Instructions field.

Additional prompts: Special Instructions and Current Message Prompt, each with Insert Condition and Verify.

Special Instructions¶

Special / additional instructions to be used in the agent. These are typically set by the Co-Admin role but can also be set by Owner, Admin, Workspace Admin, and Co-Admin.

Use {special_instructions} as a placeholder in your Agent Instructions to control where special instructions are inserted.

Current Message Prompt¶

A secondary mechanism to steer the agent's responses, appending or prepending instructions to each user prompt message.

• To create a dynamic prompt, use {message} as a placeholder in your text
• If {message} is not included, the current message is appended at the end automatically

Conditional Instructions¶

Conditional Instructions apply different instruction blocks based on tag criteria — for example a separate instruction when the user is authenticated versus unauthenticated.

They apply to Instructions, Special Instructions, and Current Message Prompt.

• Use the Insert Condition button to add a condition block
• Use the Verify button to validate the condition syntax
• Common instructions can precede conditional blocks, reducing redundancy
• Backward compatible with existing instruction sets

Conversation Inactivity Message¶

Enable Conversation Inactivity Message and enter the message in Brain > Additional prompts. The current help text says the message is sent exactly as written when the conversation becomes inactive.

Add Instruction¶

Click + Add Instruction to add additional instruction blocks, allowing multiple instruction sets on the same agent.

Channels¶

Select the channels through which users interact with your agent. Multiple channels can be enabled simultaneously.

A Customize widget action sits alongside them — see Look & Feel.

Web App & Sidebar Security¶

Tools & Artifacts¶

Adding Tools¶

Each tool appears as a row with its type badge, a Tag selector, and edit / delete actions

Click + Add Tool and pick from the tools configured for your workspace — the picker offers a search field to filter by name. Each tool appears as a row showing its name and type.

The + Add Tool picker — search, then click a tool to attach it to the agent

Tool types you will see include:

For creating and configuring tools, see Settings > Tools.

Scope Limiting (Tag-Based)¶

Every tool row carries a Tag selector — "Limit the scope to conversations with a specific tag." For example, make "Transfer to Human" available only when the tag needs_escalation is present.

# when using [name of retrieval tool] the value of query attribute must be in [language].

Artifacts¶

An Artifact is a per-agent capability that lets the agent generate a finished document directly from a chat turn. Artifacts are defined at the Global Level, made available to a workspace, and then enabled per agent. In the Artifacts group, click PDF, PowerPoint, Spreadsheet or Word Document to toggle it on or off.

Supported artifact types: PDF, PowerPoint, Spreadsheet, Word Document.

Model-agnostic: Works across Anthropic Claude, Google Gemini, and OpenAI ChatGPT model families.

For the full feature reference, see Artifacts.

Voice¶

Speech Models¶

Realtime Voice¶

For realtime model configuration, including xAI Grok Voice, see Realtime Voice.

Prompt Shortcuts¶

Prompt Shortcuts are predefined prompts displayed as buttons in the end-user chat interface. Only the label is visible to the end user — not the prompt behind it.

Shortcut Configuration Fields¶

When to Display This Prompt¶

User Defined Shortcuts¶

Turn the switch on or off. End users can create their own shortcuts for messages they use frequently.

How end users create shortcuts:

1. Within the chat interface, hover over or select a message from chat history
2. A + icon appears on the right side of the message
3. Click or tap and hold the icon to create a shortcut
4. The agent auto-generates a short label based on the message
5. The shortcut appears at the end of the chat area

Managing user shortcuts:

• Edit: Click and hold the shortcut to edit the label inline
• Delete: Click the trash can icon that appears when editing
• Mobile: Long press on a message to create a shortcut

AI Generated Shortcuts¶

Turn the switch on or off. Uses your agent's AI model to create dynamic shortcuts based on conversation context.

Available template variables:

Initial Session Tagging¶

A system-level tag internal_no_message is automatically applied at conversation start and removed on the first user message. Use it to trigger initial prompt shortcuts before a user has typed anything — for example, a welcome menu of options.

Pre Processors¶

Pre Processors run before the agent handles the user query. Script Pre Processor can set tags using JavaScript; OpenAI Moderation and Prompt Classifier provide other processing options. Configure the selected type in Settings before attaching it here.

Workflow:

1. User sends a message
2. Pre Processor executes (JavaScript snippet)
3. Pre Processor returns tags based on logic
4. Tags determine agent behavior, shortcuts visibility, tool availability
5. Query is sent to the LLM with tag context

Steps:

1. Configure the pre processor in Settings (see Settings > Processors)
2. Click + Add Pre Processor
3. Select the configured pre processor from the list
4. Review the attached processor and any displayed tags. Tag-generation logic belongs to its configuration in Settings.

Static Agent Tags¶

Static tags are predefined labels set by agent administrators to customize agent behavior and appearance without modifying the underlying template. These tags can enable or disable specific features.

// preprocessor example
const subscriberid = ld.get(data, 'user_data.subscriberid', null);
if (typeof subscriberid === "string" && subscriberid.trim().length > 0) {
  ctx.addTag('authenticated');
} else {
  ctx.addTag('unauthenticated');
}

Post Processors¶

Post Processors analyze agent responses by sending them to an LLM for evaluation, quality checks, or additional processing. Click + Add Post Processor to attach one.

Turn Post Processor¶

Executes after each user-agent exchange, analyzing individual responses in real time.

Use cases: sentiment analysis per response, compliance checking, quality scoring, real-time flagging.

Conversation Post Processor¶

Executes after the conversation goes idle, triggered by the Conversation Inactivity Duration.

Use cases: overall conversation summary, CSAT prediction, conversation categorization, post-conversation analytics.

See Settings > Processors for configuration.

Knowledge Bases¶

Context Types¶

Choose a type under Retrieval > Context:

Use Connected Knowledge Bases > Add Knowledge Base to attach bases the agent can search.

Context Options¶

Behavior Flags¶

Debug and Development¶

Language Model Control¶

User Experience¶

User Capabilities¶

When Upload to Knowledge Bases is enabled, configure Allowed Knowledge Bases — one or more knowledge bases where users can upload AI responses. Only knowledge bases where users have write access are available at runtime.

Context and Title Generation¶

Conversation Management¶

Security and Encryption¶

Encrypt Conversation — only users who participated in the conversation are able to see the message in clear text. Supervisors monitoring conversations see encrypted values.

Analytics¶

GTM Id (Google Tag Manager Id) — enable analytics data transmission and website insights. Tracks agent events like widget open, message sent, AI response received, new conversation created, and sharing of conversation transcripts.

Email Notification¶

Administrators can customize notification email templates from Settings > Templates > Email Template for both "Conversation Flag" and "Message Dislike" notifications.

Advanced Model Params¶

Chat History¶

The summary step runs once per turn that crosses the history token threshold, folding the dropped messages into the conversation's rolling summary. Default restores the built-in wording. Keep {max_summary_tokens} where you want the summary size cap — it is replaced with the live value on every call.

Retrieval¶

Large Context Processing Algorithm¶

Required. Controls how the agent handles context that exceeds the allowed token count.

Embedding Model — required when the Embeddings algorithm is selected.

Widget¶

For visual customization, see Look & Feel.

Floating Widget Auth¶

Choose Authentication Mode:

Documentation¶

Title and rich-text Content on a staging example agent. The example content describes its own fictional product.

Users open this content from the documentation button in the Web App header. Its label follows the configured Title; this example uses About this Agent.

Title — the title displayed above the documentation. For example "Quick Start Guide", "Agent Capabilities", "How to Use".

Content — rich text. Recommended content:

1. Agent Purpose:

This agent helps you find information about our products,
troubleshoot common issues, and connect with support.

2. What It Can Do:

✓ Answer product questions
✓ Provide troubleshooting steps
✓ Search documentation
✓ Escalate to human agents

3. How It Works:

1. Ask your question in natural language
2. The agent will search our knowledge base
3. Receive instant answers with source links
4. Use shortcuts for common tasks

4. Best Practices:

• Be specific in your questions
• Use shortcuts for faster results
• Provide error messages when troubleshooting
• Rate responses to help improve the agent

5. Limitations:

This agent cannot:
- Process refunds (contact support@company.com)
- Modify account settings directly
- Provide pricing quotes (contact sales)

Welcome Screen HTML¶

Replaces the blank state shown on a new chat with a configurable Welcome Screen. Admins author the landing experience — a greeting, a short description of the agent's purpose, documentation, suggested starter questions, or anything else users should see first. Applies to Web App and Sidebar modalities only; leaving it empty preserves the default empty state.

The ? icon next to the field documents allowed tags and provides sample code.

Author Affordances¶

End-User Experience¶

• Saved HTML renders as the initial state of every new chat on Web App and Sidebar, above the message input
• Clickable links or starter questions behave as the admin defined them
• Leaving the field empty preserves the existing default empty state

Agent Parameters¶

Agent Parameters are dynamic key-value variables referenced in instructions using {botParams.<key>} syntax.

Step 1: Reference a parameter in a prompt

Add a placeholder such as {botParams.company_name} to Instructions or Special Instructions:

You are a support agent for {botParams.company_name}.

Step 2: Set its value

Add Agent Parameters through Add a section if needed. The company_name key appears automatically from the prompt. Enter its Value, for example fifthelement.ai, then save the agent when the configuration is ready.

Benefits:

• Easy updates without touching instructions
• Reusable agent configurations
• Multi-environment configuration (dev, staging, prod)

Translation¶

Search Filters¶

Search Filters control which documents a search returns. Click Add rule to create a rule. Set its Label, Vector Store Key and Type (Select Box, Text, Number Range, URL or Date). Use Use in Filters and Show in Results to control where the field appears.

Related Topics¶

• Agent Builder - Basic Configuration - Essential setup steps
• Agent Builder - Full reference for every section and field
• Agent Reports - Monitor agent performance and analytics
• Agent's Look & Feel - Customize branding and appearance
• Cloning an Agent - Duplicate and modify existing agents
• Version Control - Manage agent versions and rollback
• Knowledge Overview - Configure knowledge bases for RAG
• Settings > AI Models - Manage available language models
• Settings > Processors - Configure pre/post processors