Agents

Learn how to create and configure AI agents step by step using the agent builder.

Creating Your First Agent

Step 1: Profile & Instructions

Configure your agent’s basic information and behavior.

Avatar

Upload a custom image or choose from preset avatars
Click on the avatar selector to browse options

Agent Name (Required)

Enter a descriptive name for your agent
This will be visible to users interacting with the agent

Description (Required)

Write a clear description of what your agent does
Explain how it can help users (minimum 4 rows)
This description appears when users interact with your agent

System Instructions

Write detailed instructions that define your agent’s role and behavior
Use the markdown editor to format your instructions
Click “Enhance with AI” to improve your instructions automatically
Instructions guide how the agent interacts and responds to users

Step 2: Capabilities

Configure the AI model, tools, and integrations your agent can use.

Screenshot: Capabilities Tab

Model Selection

Choose the LLM model to power your agent
Available models include GPT-4, Claude, and others
Different models have different capabilities and costs

Model Settings Click “Configure” to adjust:

Temperature: Control response creativity (0-2)
- Lower = more focused and deterministic
- Higher = more creative and varied
Max Tokens: Set maximum response length
Top P: Nucleus sampling parameter
Frequency Penalty: Reduce repetition
Presence Penalty: Encourage topic diversity

Tools Enable tools your agent can use:

Core tools (planning, file operations, user interaction)
Search & research capabilities
Code execution in sandbox
Data analysis and visualization
PDF processing, email, image generation

Integrations

Click “Browse Apps” to connect third-party applications
Over 1000+ apps available through Composio
Select specific actions your agent can perform
Click “Custom MCP” for custom integrations

Triggers

Click “Browse Triggers” to set up automatic agent activation
Configure event-based, scheduled, or webhook triggers
Triggers activate your agent automatically based on conditions

Step 3: Knowledge

Connect knowledge bases your agent can access for answering questions.

Agent Knowledge

Adding Knowledge

Click “Add Knowledge” to browse available knowledge bases
Use the search bar to find specific knowledge sources
Select checkbox next to each knowledge base to include
View knowledge type and chunk count for each source
Click “Confirm Selection” to add selected knowledge

Managing Knowledge

View all connected knowledge bases in the list
Each knowledge base shows:
- Knowledge type (File, Website, Database)
- Number of content chunks
Click the trash icon to remove a knowledge base

Step 4: Settings

Configure advanced features and structured output.

Agent Knowledge

Agent Features

Enable or disable specific capabilities
Toggle features on/off using switches
Each feature has a description of its functionality

JSON Schema (Optional)

Define structured input/output format for your agent
Use the code editor to write JSON schema
Click “Enhance with AI” for schema suggestions
Useful for API integrations and structured data

Make your agent available and share it with others.

Agent Publish

Publishing

Toggle “Published” switch to make agent publicly available
Published agents can be accessed via URL or embedded
Unpublished agents are only available in your workspace

Published URL

Copy the agent URL to share with others
URL includes authentication for secure access
Anyone with the URL can interact with your agent

Embed Code

Choose embed type: iframe or script
Copy embed code to add agent to your website
Available in two modes:
- Full Window: Agent takes entire page
- Bubble: Floating chat bubble in corner

Channel Integrations Configure agent deployment to messaging platforms:

WhatsApp: Connect via WhatsApp Business API
Slack: Deploy agent to Slack workspace
Twilio: Integrate with Twilio messaging

Testing Your Agent

Agent Testing

Right-Side Chat Panel

Save your agent first to enable testing
Use the built-in chat interface to test
Send messages and verify agent responses
Ensure tools and integrations work correctly
Test with different types of questions

Saving Your Agent

Agent Testing

Save Options

Click “Save Agent” button at the bottom
“Unsaved changes” indicator shows when edits are pending
Changes must be saved before testing
All tabs save together as one agent configuration

Cancel Option

Click “Cancel” to discard changes
Confirmation dialog appears if there are unsaved changes
Returns to agent list without saving

Managing Agents

Editing

Open an existing agent to edit
Make changes in any tab
Save to update the agent configuration

Deleting

Click “Delete” button (only for existing agents)
Confirmation dialog prevents accidental deletion
Deletes agent configuration permanently
Chat history is preserved but agent cannot be used

Best Practices