Python familiarity is required. If you are a non-technical operator, see Configure agents in the UI for the no-code path, or share this page with your developer.
What you can build
API integrations
Call external APIs to look up bookings, check availability, or retrieve account details
Business logic
Validate input, calculate pricing, or enforce rules that the LLM cannot handle alone
CRM and data writes
Create records, update customer profiles, or log interactions in external systems
Dynamic routing
Route calls to the right team based on real-time data or custom logic
Multi-step workflows
Build code-driven flows with transition functions for complex conversation paths
Real-time configuration
Define schemas that adapt agent behavior based on external config data
Get started
Write your first function
Follow the create a function guide to set up a Python function with parameters, code, and a description the LLM can use to trigger it.
Understand the tools system
Read the tools overview to learn how functions integrate with the LLM, how to use
conv objects, and best practices for naming and triggering.Explore advanced patterns
Dive into flows, call handoffs, and secrets management to build production-ready integrations.
Tools and functions
Write Python scripts that run when your agent needs to take action. Functions are the bridge between conversation and your business systems.Create a function
Set up a function with naming conventions, parameters, and Python code
Start function
Initialize conversation context before the greeting plays
End function
Run post-call processing after a conversation ends
Return values
Control agent behavior with string and dictionary returns
Variables
Define, update, and persist values across turns
Libraries
Standard and non-standard libraries available in your functions
Delay control
Add filler phrases to avoid silence during slow functions
Use functions in topics
Trigger functions from managed topics
Reference
Access conversation data and built-in utilities from within your functions.conv object
Conversation states, flows, and telephony attributes
conv.log
Structured diagnostics and PII-scoped logging
conv.utils
Built-in helpers for addresses, cities, and structured data
conv.api
Call configured API integrations
Agent memory
Persistent data across conversations for repeat callers
Function classes
VoiceWeighting, TTSVoice, and provider classes
Advanced capabilities
Secrets
Store API keys and credentials securely
Call handoffs
Configure call transfers and routing rules
Configuration builder
Define real-time config schemas for dynamic agent behavior
Call data export
Access transcripts and recordings via API
Code-driven flows
Build flows with transition functions and programmatic control
API reference
Integrate with PolyAI REST APIs for chat, conversations, alerts, and more