> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://poly-ai.ferndocs.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://poly-ai.ferndocs.com/_mcp/server.

# 9.2025

> September 2025 release notes.

The **September 2025** PolyAI Agent Studio release focuses on the addition of new Webchat configuration options and individual Test Cases
to complement the previously introduced Test Sets.

Expand the items for details:

#### Webchat configuration

Configure, preview, and test non-voice Webchat deployments directly in Studio.

![webchat-config](/_fern-files/_img/5e3ec13e0a3e0a187a0cf18fdd456d00d4077164d3bd949a444ffb4c1b07d4f9.webp)

**What's new:**

* There is a new **Webchat** configuration section in the Agent Studio **Settings** tab.
* You can customize the appearance of the chat widget quickly using the live preview.
* You can test the widget quickly in-browser with **Preview Demo** or in any site using the embed script tag:

![one-webchat-click](/_fern-files/_img/3270b61e3e4fc36c367c8ffe8f794f459ce2b64cf71077a7294023c21c2f7d4a.webp)

#### Test cases

The Test Suite now supports individual **Test Cases**.

![test-cases](/_fern-files/_img/f223c11a32e159ef755bc86bf71bf3ce616b04ae6172e41353d63bf3c2e570cb.webp)

**What's new:**

* You can use a conversation to create a **Test Case** against a single scenario.
* These are run directly from the **Test Cases tab**, after selecting the agent version (Draft/Sandbox) to test against.
* Outcomes and last-run timestamps are tracked automatically.
* Organize cases into **Test Sets** to rerun whole groups at once and view aggregated results with charts.
* Cases can belong to **multiple sets**, so you can target them by feature or however else is required.

#### Home page updates

The Agent Studio home page now surfaces more trend data and a clearer quality summary.

![home-v2](/_fern-files/_img/1815a2e7b943decf80296540cf3c3baf93323a9921dd40b427b7c56544e7bdd1.webp)

**What's new:**

* **Average PolyScore** added to Quick Insights
* New charts for **Containment rate** and **PolyScore** over time
* Consistent "**PolyScore**" naming across tables, filters, and details

#### Smart Analyst: reasoning logic and chat history

Smart Analyst now supports structured responses using reasoning logic. A persistent chat history has also been added.

![smart-analyst-reasoning](/_fern-files/_img/12930331574497be32f74042872c2fbf1f04be03020b29e8c8d4e44d8ab2975a.webp)

**What's new:**

* **Structured responses**: answers now include reasoning logic along with the final answer, as well as more clearly labeled sources and steps.

![smart-analyst-history](/_fern-files/_img/b6ec41084449d628c8554fb93164bd6edccbae32df6a28a3c349d1a31cdedf45.webp)

* **Chat history**: resume previous analyses; conversations stored with redacted transcript data
* **Auto titles** with the ability to rename and delete old analyses from the history list.

#### Numbers tab

You can now buy and assign phone numbers from the new **Numbers** tab in the Agent Studio UI, and you can optionally link these to individual multi-site **variants**.

![phone-numbers](/_fern-files/_img/e92b04cb8a58835f1a00906401b81ceeea32ddbf4d2f285c2b7f0e193e242442.webp)

**What's new:**

* New **Numbers** page under *Configure* shows all numbers by environment
* Add, view, and delete numbers in Sandbox and Pre-release.
* Assign numbers to variants for multi-site projects.

#### Function logging (conv.log)

Functions can now write structured logs [Conversation Review](/agent-studio/conversations/annotations) and the [Conversations API](/api-reference/conversations) using [`conv.log`](/agent-studio/tools/classes-and-objects/log).

**What's new:**

* Use `conv.log.info`, `conv.log.warning`, and `conv.log.error` inside functions
* Entries appear in **Conversation Review → Diagnosis** and using the **Conversations API**.
* Mark entries as **PII** so they are only visible to users with permission.

#### Conversations API: tool calls

Third-party QA tools and automated test suites can now read **tool/function call events** from the Conversations API.

**What's new:**

* Every function call made during a conversation is now surfaced in the API response
* Makes it possible to verify not just the dialogue, but also the actions taken behind the scenes
* Access is scoped per environment (Sandbox, Pre-release, Live) so external tools don't touch production data