What Are AI Agents?
Agents are the core building blocks of Parallel AI. Each agent is a fully configurable AI assistant that you can deploy across multiple channels — website widget, email, SMS, and voice — to engage your customers automatically. Agents can be equipped with:- Custom system prompts to define personality, tone, and purpose
- Knowledge base access to answer questions from your uploaded documents
- Multi-step tool calling to chain multiple actions in a single conversation turn
- Lead capture to passively extract and store contact information from conversations
- Channel-specific settings for fine-grained control per deployment surface
Agent Configuration Fields
Channel Types
Agents can be deployed across four channel types. Each is configured independently underchannelSettings.
Website Widget
Embed a chat widget on any webpage. Customize colors, avatar, and welcome message to match your brand.
Connect an email address so the agent reads and responds to inbound emails automatically.
SMS
Attach a phone number to your agent and handle text message conversations at scale.
Voice
Deploy the agent on a voice channel for automated phone call handling.
channelSettings object:
Creating an Agent
1
Define your agent's purpose
Write a clear
systemPrompt that describes what the agent should do, its tone, and any constraints (e.g., “Only answer questions about our product.”).2
Choose your channels
Enable the channels your customers use. You can always add more later by updating the agent.
3
Enable the knowledge base (optional)
If your agent needs to answer questions from documents, set
knowledgeBaseEnabled: true and upload documents via the Documents feature.4
Create via API
Send a
POST /api/v0/agents request with your configuration.Shareable Agent URL
Every agent is automatically assigned ashortCode upon creation. Use it to build a public-facing URL for your agent:
Multi-Step Tool Calling
WhenmultiStepToolsEnabled is true, your agent can chain tool calls across multiple turns in a single conversation exchange. This is useful for complex workflows where the agent needs to look up information, process a result, and then take a follow-up action.
Use multiStepMaxTurns to cap the number of chained turns (between 1 and 10, default 5) to control cost and latency.
AI Contact Capture
WhenaiContactCaptureEnabled is true, the agent passively monitors conversations for contact information (name, email, phone number, etc.) and automatically saves it to the linked Smart List defined by contactListId.
No explicit form or prompt is needed — the agent extracts contact details naturally as they appear in conversation.
To store captured leads, set
contactListId to the ID of an existing Smart List. If no list is linked, captured data will not be persisted.Next Steps
Agents API Reference
Explore the full list, create, update, and delete endpoints for agents.
Knowledge Base Documents
Upload documents to power your agent’s knowledge base.
Smart Lists
Set up contact lists to store leads captured by your agents.
Sequences
Build automated outreach campaigns for captured leads.
