06.2025 - PolyAI Platform

Version comparison enhancements

You can now compare versions across functions, flows, and Knowledge items–not just KBs.

What’s new:

The sidebar UI highlights additions, edits, and deletions
View full publish history per item

This update supports better change management for teams working across large or complex agents, especially when multiple people are working on a single item.

Agent configuration

Bring dynamic, real-time settings to your agent–without touching flows or code.

How it works:

Builders define a JSON schema to create structured config forms
Managers populate values per environment (Draft, Pre-release, Live)
Forms appear in the Configuration tab once the schema is published

This setup lets teams safely update logic–like hours, routing, or toggles–without risk to underlying functions.Example uses:

Opening hours
After-hours message toggles
Fallback numbers
Environment-specific logic flags

Action name search and autofill

You can now invoke actions, like functions and SMS, by typing /name directly into the editor.Improvements include:

Cleaner UI for selecting actions
Matches top-level categories and partial names
Can better handle nested actions in the UI

Cached disclaimer messages

Disclaimer messages are now cached in the Cache under Audio Management.

Why this helps:

Ensures consistent delivery of disclaimer audio
Reduces response time when disclaimers are reused across flows or agents
Simplifies troubleshooting when disclaimers fail to play

Smart analyst (Beta)

This feature is currently in open beta. Reach out to your PolyAI contact for access and setup.

Smart Analyst is a new chatbot in Studio that reviews a random sample of your recent calls and answers natural-language questions about them.Ask questions like:

“How long are my calls on average?”
“Which intents failed yesterday?”
“What’s the most common escalation reason?”

How it works:

Reviews call data from the past 24 hours
Summarizes behavior patterns and performance
Lets you explore agent usage conversationally

Agent memory

Agent Memory allows agents to persist and recall information across conversations–like user preferences or recent bookings–using a secure key-value store.How it works:

Memory is stored per identifier (e.g. phone number)
Functions can access memory using conv.memory
Values are written using conv.state and saved at call end
Multiple identifiers (like phone and email) can be linked to the same profile

This is especially useful for repeat callers or omnichannel use cases, where remembering previous interactions improves containment and personalization.Example use cases:

Recall a user’s previous delivery date or cheese preference
Link web and phone interactions to a shared memory profile
Personalise greetings based on previous contact history

Developer usage:

# Read memory
cheese = conv.memory.get("cheese_type")

# Persist memory
conv.state["cheese_type"] = "gouda"

Before using Agent Memory in production, a Data Protection Impact Assessment (DPIA) may be required. Contact your PolyAI representative for guidance on compliance requirements.

Test Suite

The Test Suite lets you save real conversations as test cases and re-run them against draft or sandbox versions to check for behavioral drift.What’s new:

Save test cases from the chat panel or conversation review
Re-run conversations with new parameters like contact name or phone number
Run tests individually or batch test cases in groups
Review success rates in the Test Runs tab

Use this to catch issues when refining agents, especially when upgrading functions or dealing with variants.

PolyScore

PolyScore introduces automated LLM-based scoring for conversations–directly in Conversation Review.What it does:

Assigns a score from 0–10 per call, where higher is better
Weighs multiple behavioral and qualitative signals
Offers fast, standardized feedback across all calls

PolyScore also appears as a filterable field in the Conversations tab to help teams quickly identify high- or low-performing calls.

Documentation Index