Skip to content

Agent Builder

Complete reference for every section and field in the Agent Builder.

Overview

The Agent Builder is a modular page. It opens with a mandatory core — Identity and Brain — and every other capability is a self-contained module you add on demand.

This page is the full field-by-field reference. For a guided walkthrough, start with Basic Configuration; for detail on each module, see Advanced Configuration.

Modular Agent Builder

The Builder opens with Identity and Brain only — everything else is added from Add a section

Prerequisites

  • Access to https://PLATFORM-URL-PLACEHOLDER
  • Permission to create or edit agents
  • Configure shared models, tools and processors in Settings. Manage Knowledge Bases under Knowledge.

Page Anatomy

Element Description
Agent name Inline-editable heading at the top of the page. Reads Untitled Agent until set
Mode chip Chat Mode or Search Mode
Change Mode Returns to the mode picker. Shown while creating an agent
Save state Shows Not saved yet, an unsaved-change count, or All saved
Agent sections sidebar Lists sections on the left and navigates to them
Add a section Opens the searchable section chooser
Create Agent / Save Sticky header action for creating or saving an agent
Back to Agent Returns to the agent, when editing an existing one
Compare Version Opens version comparison. See Version Control

In the sidebar, Identity and Brain carry a lock icon and cannot be removed. Optional sections have reorder handles: drag them, or focus a handle and use the arrow keys. Open a section's actions menu to remove it.

Agent Modes

Agent creation opens with a mode choice: "What kind of Agent are you building?"

Mode Description
Chat A conversational Agent that answers questions and helps users in real time
Search Retrieves answers from your knowledge first, then summarises with citations

The mode decides which modules and defaults are pre-loaded.

Chat Search
Search Mode selector in Identity — Yes
Search Filters section — On the page by default
Capabilities modules offered Tools & Artifacts, Voice, Prompt Shortcuts, Pre Processors, Post Processors Tools, Pre Processors, Post Processors
Knowledge modules offered Knowledge Bases Search Filters
Distribution modules offered Channels Channels
Tuning modules offered Behavior Flags, Advanced Model Params, Agent Parameters, Documentation Behavior Flags, Advanced Model Params, Agent Parameters, Documentation

Identity

Mandatory. "Who is this Agent?"

Field Required Description
Agent name Yes Set inline at the top of the page. Internal identifier used in the platform, logs, and management
Description Yes One line summary of what this Agent does
AI Name No Shown to end users in chat (e.g. "AI", "Astra", "Helper"). Defaults to AI
Greeting No Shown as the first message in the chat window
Notes No Internal notes, not shown to users
Agent Icon No Icon displayed before the agent's name in the chat interface
Template No Workspace Public or Unlisted (default)
Agent Categories No Groups the agent in admin lists without changing visibility or behavior
Remarks — Read-only provenance line shown when editing an existing agent — for example, that it was cloned from another agent with its source name, ID and version
Search Mode Search agents only Search, AI, or User Selectable (default). User Selectable lets the end user toggle between AI and Search mode

Brain

Mandatory. "How does it think?"

Instructions

Field Description
Instructions The system instruction steering every reply — how the Agent should behave, its tone, what it should refuse
Special Instructions Special / additional instructions. Referenced via the {special_instructions} placeholder
Current Message Prompt Appended or prepended to each user message. Use {message} as a placeholder; if omitted, the current message is appended at the end
Conversation Inactivity Message When enabled, sends the configured message exactly as written when the conversation becomes inactive, according to the current field help
+ Add Instruction Adds further instruction blocks

Controls available on the instruction fields:

Control What it does
Default Restores the built-in wording
Generate with AI Drafts a full instruction set from a short prompt, for review before applying
Insert Condition Adds a tag-based conditional instruction block
Verify Validates the condition syntax

Placeholders {special_instructions} and {user_instructions} are substituted at runtime.

Available User Data

Use {{data.userInformation.email}}, {{data.userInformation.name}} and {{data.userInformation.id}} in prompts to personalize responses. Available for logged-in users on Web App and Sidebar. For dynamic user data parameters, use {{data.keyname}}.

Models

Field Required Description
Available Language Models Yes One or more models. Each row carries Default / Set Default, a Tag selector limiting that model to conversations with a specific tag, an edit (pencil) icon opening Modify Language Model, and a delete icon

Modify Language Model (per-model)

The pencil icon on a model row opens Modify Language Model — parameters that apply to that one model on this agent.

Field Description
Language Model The workspace model this row points at
Add Tag Tag that scopes this model
Display Name Name shown for this model
Description Free-text note
Temperature Slider controlling randomness
Max Output Tokens Cap on output. The field shows the model's maximum and a recommended value
Per 1k Tokens Price ($) Used for live cost display. Requires Display Session Cost under Behavior Flags
Max Messages in History How many prior messages are passed. The field shows a per-model recommendation
History Tokens Threshold Token count past which the history strategy kicks in. The field shows a per-model recommendation

Three places models are configured

Keep the levels straight:

Level Where What lives there
Workspace Settings → Language Models Model name, credential, provider model, Thinking Level
Per model, per agent Brain → model row → pencil The Modify Language Model fields above
Agent-wide Advanced Model Params Use Chat History, summarization, condense-question, Large Context Processing Algorithm

Voice

Language Models are not used during Realtime Voice interactions.

Modules

Click Add a section in the left sidebar, then search or browse the available sections.

Group Module Purpose
CAPABILITIES Tools & Artifacts Retrieval, API calls, static text, custom logic, document generation
Voice ASR, TTS and realtime speech-to-speech
Prompt Shortcuts Predefined and AI-generated quick actions
Pre Processors JavaScript that runs before the LLM, setting tags
Post Processors LLM evaluation after a turn or a conversation
KNOWLEDGE Knowledge Bases Text, documents and knowledge repositories
DISTRIBUTION Channels Web App, Sidebar, Floating Widget and sign-in rules
TUNING Behavior Flags Debugging, UX toggles, encryption, analytics, email notifications
Advanced Model Params Chat history, summarization, large-context handling
Agent Parameters Reusable {botParams.<key>} variables
Documentation User-facing guide and Welcome Screen HTML

The chooser also includes these sections. Widget and Floating Widget Auth require Floating Widget to be enabled in Channels; Translation can be added separately:

Module Purpose
Widget New-conversation button, error message, allowed domains
Floating Widget Auth No Authentication / Required Login / Optional Login
Translation Translation toggle, model and prompt

Search agents additionally carry Search Filters.

Module Field Summary

Channels

Field Description
Web App Access the agent directly by typing its address
Sidebar Use the agent as a sidebar to an internal or external website
Floating Widget Classic website agent for assisting visitors
Customize widget Opens widget appearance settings
Web App & Sidebar Security Sign-In Required or Sign-In Optional

Tools & Artifacts

Field Description
+ Add Tool Attaches a tool configured in Settings
Tag (per tool) Limits the scope to conversations with a specific tag
Artifacts Toggles document-generation artifacts on or off for this agent

Tool types include Retrieval, Context, Playbook, Webhook, Custom, User Memory and Data Visualization.

Voice

Field Description
Push To Talk Hold to speak rather than streaming continuously
Speech Recognition Model (Input) ASR model
Text To Speech Model (Output) TTS model
Realtime Voice Processing Enable / Disable
Launch in Voice Mode by Default Enable / Disable. Default Disable. Opens the agent straight on the voice screen
Realtime Voice Model (Includes AI Listens) Required. The realtime model
SIP Realtime Voice Processing Enable / Disable. Telephony
Websocket Realtime Voice Processing Enable / Disable. Browser-based voice

Prompt Shortcuts

Field Description
+ Add Shortcut Label, prompt/hyperlink, one-off tag, hyperlink flag and chat-history flag
User Defined Shortcuts Allow / Disallow
AI Generated Shortcuts Allow / Disallow
Number of AI Shortcuts 1 to 5. Default 3
AI Shortcut Generation Template Prompt template with a live token count

Pre Processors

Field Description
+ Add Pre Processor Attaches a processor configured in Settings
Static Agent Tags Predefined labels that customize agent behavior and appearance

Post Processors

Field Description
+ Add Post Processor Turn Post Processor or Conversation Post Processor

Knowledge Bases

Field Description
Limit Context Token / Context Token Limit Caps context size. If available tokens are fewer than the limit, available tokens are used
Use Dynamic Context from the Webpage For Floating Widget / Sidebar agents
+ Add Context Text, Document, or Knowledge Base

Warning

Add the Context Tool under Tools & Artifacts to enable context usage with the agent.

Behavior Flags

Grouped into Debug and Development, Language Model Control, User Experience, User Capabilities, Context and Title Generation, Conversation Management, Security and Encryption, Analytics, and Email Notification. See Behavior Flags for every toggle.

Advanced Model Params

Field Description
Use Chat History Pass prior messages to the model on each turn
History Summary Strategy Truncate (default) or Summarize
Summarizer Model Required when summarizing
History Summary Instructions Prompt maintaining the rolling summary. Keep {max_summary_tokens}
Condense Question Before Retrieval Rewrites the question into a standalone query
Condense-Question Model Required when condensing
Large Context Processing Algorithm Required. Truncate, LLM Prompt Chaining, or Embeddings
Embedding Model Required for the Embeddings algorithm

Agent Parameters

Key-value pairs referenced in instructions as {botParams.<key>}.

Documentation

Field Description
Title Title displayed above the documentation
Content Rich text guide opened from the agent documentation button; its label follows the configured Title
Welcome Screen HTML Initial state of a new chat on Web App and Sidebar. Supports Preview, Use template, Generate with AI and Revert

Widget

Field Description
Allow Users to Start a New Conversation Shows a "New conversation" button in the widget toolbar
Error Message Shown when the Agent cannot respond
Allowed Domains Comma-separated hostnames. Blank allows all origins

Floating Widget Auth

No Authentication, Required Login, or Optional Login.

Translation

Field Description
Translation Enable / Disable
Translation Language Model Required. Model used for translation tasks
Translation Prompt Prompt used for translation tasks

Search Filters

Search agents only. Define search filter rules with Add rule.

Warning

To use the Vector Store Key dropdown, first select a retrieval tool in the Tools section.

Creating and Saving

Use Create Agent in the sticky header while creating, or Save when editing. The header shows Not saved yet, an unsaved-change count, or All saved.