Agent Builder
Overview
The Agent Builder is the visual workspace where you design and configure AI agents. It combines identity settings, skill assignment, data resource connections, personality tuning, and a visual task canvas — all in one interface.
Purpose
The Agent Builder lets you create highly customized AI agents by:
- Defining identity — name, role, bio, icon, and accent color
- Assigning skills — modular capability packs that give the agent specific expertise
- Connecting data resources — link integrations (Google Drive, Slack, BigQuery, etc.) so the agent can access your data
- Tuning personality — set the tone, style, and verbosity of the agent's responses
- Writing a system prompt — provide custom instructions that guide the agent's behavior
- Building task workflows — create visual multi-step task flows on a drag-and-drop canvas
How It Works
Builder Layout
Agent Builder
├── Top Bar
│ ├── Back to Hub
│ ├── Agent Name (editable)
│ └── Save / Deploy Button
├── Left Sidebar (collapsible panels)
│ ├── Identity Panel
│ ├── Skills Panel
│ ├── Data Resources Panel
│ ├── Personality Panel
│ └── System Prompt Panel
└── Main Canvas
└── Task Workflow (visual node editor)
Identity Configuration
The Identity panel lets you set the agent's core attributes:
| Field | Description |
|---|---|
| Name | Display name shown in the Agent Hub and chat |
| Role Title | Descriptive role (e.g., "Data Analyst", "Marketing Strategist") |
| Bio | Short description of the agent's purpose and capabilities |
| Icon | Choose from a library of icons to represent the agent |
| Accent Color | Pick from 8 preset colors for visual branding |
Skills Panel
Skills are modular capability packs that give your agent specific expertise. See Agent Skills & Tools for full details.
From the Skills panel you can:
- Browse available skills — see all prebuilt and custom skills
- Assign skills — click to add a skill to the agent
- Remove skills — unassign skills the agent no longer needs
- View skill details — see each skill's triggers, tool categories, and description
Each skill has:
- A display name and description
- Triggers — keywords that activate the skill when matched in user messages
- Tool categories — which platform operations the skill unlocks (Dashboard, Workflow, AI, Integration)
Data Resources Panel
Data resources connect your agent to external services and data stores. When a data resource is linked, the agent can search, browse, and retrieve information from that service.
Supported data resources include:
| Resource Type | Capabilities |
|---|---|
| Google Drive | Browse folders, search files, preview documents |
| OneDrive | Browse folders, search files |
| BigQuery | Explore datasets, list tables, preview schemas |
| Gmail | Search emails, list labels |
| Slack | Search messages and channels |
| Jira | Search issues and projects |
| Asana | List workspaces and tasks |
| Shopify | Search products, orders, customers |
| QuickBooks | Search invoices, customers, accounts |
| Databases | Explore tables, run queries (PostgreSQL, MySQL, etc.) |
| Calendar | List events from Google Calendar or Outlook |
Note: Data resources use your existing integration connections. You must first connect the integration in Settings → Integrations before it appears as an available data resource.
Personality Settings
Fine-tune how the agent communicates:
| Setting | Options | Description |
|---|---|---|
| Tone | Professional, Friendly, Casual, Formal | The emotional register of responses |
| Style | Concise, Detailed, Conversational, Technical | How responses are structured |
| Verbosity | Brief, Moderate, Thorough | How much detail is included |
System Prompt Editor
The system prompt is the core instruction set that defines the agent's behavior. Write free-form instructions to:
- Define the agent's expertise domain
- Set boundaries on what it should and shouldn't do
- Provide context about your organization or use case
- Include specific formatting preferences
The system prompt is combined with the agent's identity, active skills, and personality settings at runtime to produce the full instruction set sent to the AI model.
Model Tier Selection
Choose the AI model tier for the agent:
| Tier | Description |
|---|---|
| Fast | Quick responses, lower cost — ideal for simple lookups and routine tasks |
| Pro | Balanced capability and speed — recommended for most agents |
| Deep Thinking | Maximum reasoning power — best for complex analysis and multi-step planning |
The model tier determines which underlying AI model is used when the agent processes messages. See Agent Teams for more on model tiers.
Task Canvas
The Task Canvas is a visual node editor for designing multi-step workflows that the agent can execute. Tasks appear as connected nodes on a drag-and-drop canvas.
Canvas features:
- Add task nodes — each node represents a discrete step the agent should perform
- Connect nodes — draw edges between nodes to define execution order and dependencies
- Configure nodes — set instructions and parameters for each task step
- Zoom & pan — navigate large task workflows with scroll and drag
Task workflows are useful for agents that need to perform complex, repeatable sequences — for example, "Fetch sales data from Shopify → Summarize trends → Create a dashboard tile → Send a Slack notification."
Settings
Saving & Deployment
| Action | Description |
|---|---|
| Save | Saves the current configuration (identity, skills, data, prompt, tasks) |
| Back to Hub | Returns to the Agent Hub; unsaved changes prompt a confirmation |
All changes are saved to the agent's configuration and take effect immediately for new conversations.
Use Cases
1. Building a Data Analyst Agent
Set the role to "Data Analyst", assign dashboard and workflow skills, connect BigQuery and Google Drive as data resources, and set the tone to "Professional" with "Detailed" style.
2. Creating a Customer Support Agent
Name it "Support Assistant", write a system prompt with your product FAQ, assign integration skills for Zendesk and Gmail, and set the verbosity to "Thorough."
3. Designing a Multi-Step Reporting Agent
Use the Task Canvas to define a workflow: pull data from BigQuery → run an AI summary → push results to a dashboard → email the report. The agent executes this sequence on demand.