What is an Agent?
An agent is your AI-powered assistant that interacts with users on your behalf. Think of it as a virtual team member that’s available 24/7. Each agent consists of:- System Prompt: Instructions that define the agent’s personality and behavior
- Knowledge Base: Custom data the agent uses to answer questions accurately
- Actions: Tasks the agent can perform (scheduling, data collection, etc.)
- Channels: Where the agent is deployed (widget, Slack, Telegram, Discord)
Agent Types
Support Agent
Answer customer questions using your knowledge base. Ideal for FAQ, product support, and troubleshooting.
Sales Agent
Qualify leads, collect contact information, and guide prospects through your sales funnel.
Booking Agent
Schedule appointments and meetings via Calendly integration.
Custom Agent
Build any conversational experience tailored to your specific needs.
Agent Components
System Prompt
The system prompt is the most important part of your agent. It defines:- Personality: How the agent communicates (friendly, professional, casual)
- Capabilities: What the agent can and cannot do
- Guidelines: Rules for handling specific situations
- Tone: The voice and style of responses
Knowledge Base
Train your agent with your own data:- Upload documents (PDF, DOCX, TXT)
- Crawl websites automatically
- Add structured FAQs
- Paste raw text content
Actions
Extend your agent’s capabilities:- Human Handoff: Transfer to live support
- Data Collection: Gather user information
- Integrations: Connect to Calendly, Shopify, etc.
Channels
Deploy your agent to multiple platforms:- Web Widget
- Slack
- Telegram
- Discord
- Phone (via Twilio)
Agent Settings
| Setting | Description | Example |
|---|---|---|
| Name | Display name for the agent | ”Customer Support” |
| Description | Internal description | ”Handles product questions” |
| System Prompt | Instructions for the AI | See examples below |
| Model | LLM provider | GPT-4, Claude, Gemini |
| Temperature | Response creativity (0-1) | 0.7 (balanced) |
| Language | Primary language | English |
How Agents Work
- User sends a message via any connected channel
- Agent processes the message and understands intent
- Knowledge base is searched for relevant information
- Context is retrieved and added to the AI prompt
- Response is generated using the LLM
- Actions are executed if triggered (scheduling, data collection)
- Response is sent back to the user
Credits Usage
Each agent interaction consumes credits from your workspace:| Action | Credits |
|---|---|
| Chat message | 1 credit |
| Voice minute | 5 credits |
| Knowledge base query | 0.5 credits |
| Action execution | 1 credit |