Configuration

Configure LLM, greeting, and safety filters for chat.

Configure chat settings in Channels > Chat. Create widgets at Widget.

General

Select the language model that powers your chat agent.

Large language model · dropdown

Choose the LLM for chat responses. Available models include PolyAI’s Raven 3.5 (recommended), OpenAI (GPT series), and Amazon Bedrock options.

Recommended for chat: PolyAI’s Raven 3.5 is designed for conversational AI and tuned for both voice and chat. It supports custom style instructions, auto-reasoning for complex queries, out-of-domain detection, and 24+ languages. Raven V3 is optimized for voice only and can only be selected in Voice configuration — Raven 3.5 extends all Raven capabilities to chat.

Style prompt

Use this prompt to tailor tone, format, and etiquette specifically for the chat channel.

Style prompt · string

An optional multiline prompt that adjusts how your agent responds in chat. Use this to set channel-specific formatting or communication style that differs from voice.

Example: “Use concise responses with bullet points. Include links when helpful.”

The style prompt is optional. If left empty, the agent uses its default persona across all channels.

Greeting

Configure what your agent says when a chat session begins.

Greeting message · string

The first message your agent sends when a user opens the chat widget.

Example: “Hello, how can I help?”

Keep your greeting short and welcoming. Avoid long introductions that push the conversation off-screen.

Safety filters

Safety filters are configured on a per-channel basis. The chat channel has its own safety filter settings, separate from the voice channel and the project-wide defaults.

Enable safety filters · boolean

When enabled, applies content filtering to chat responses. These filters apply to the chat channel only. Default settings can be managed on the Configure > General page.

When safety filters are enabled, adjust the strictness for each category using the sliders:

Category	Description
Violence	Controls filtering of violent content (Lenient → Strict)
Hate	Controls filtering of hateful or discriminatory content (Lenient → Strict)
Sexual content	Controls filtering of sexually explicit content (Lenient → Strict)
Self-harm	Controls filtering of self-harm related content (Lenient → Strict)

The jailbreak attack filter is always enabled and cannot be turned off.

More lenient settings allow a wider range of content through. Review your use case and compliance requirements when adjusting these settings.

Branching behavior

Chat configuration changes take effect immediately when saved. Unlike other project settings, webchat does not follow the environment branching system – there is no Sandbox, Pre-release, or Live promotion step. Use the Test Agent Chat panel to verify changes before saving.

Markdown support

Chat messages support formatted markdown rendering:

• Bold and italics
• Bulleted and numbered lists
• Links
• Code blocks

Formatted markdown appears in both the Test Agent Chat panel and Conversation Review.

Next steps

After configuring chat settings:

1. Configure the widget - Style and customize the chat widget appearance
2. Deploy to your website - Generate and embed the script tag

You must configure chat settings before the widget will function. In-product banners will guide you through this process.