Feature requests

We would like to have the option not to let the chatbot respond to separate emojis or images. When a customer sends a single emoji, the bot often misinterprets it. The bot also responds by default when receiving images or documents, which can be experienced as annoying or spammy by the customer. A setting to disable this behavior would be desirable.

You want the AI agent to receive more detailed feedback than just a thumbs down. This would help the agent to improve further. With CHATGPT: When you give a thumbs down on an answer, the agent might ask you to provide more context about why you made that choice. You could indicate that the answer is factually incorrect, that you don't like the style, or that the instructions were not fully followed (see photo). This type of detailed feedback would enable the AI agent to better understand exactly what went wrong, and would ultimately improve the agent's strength and effectiveness.

Have the AI Agent only provide answers based on the knowledge added in Watermelon, without relying on OpenAI’s database for additional information. This ensures that the chatbot only shares company-approved documentation and does not generate answers from external sources. This feature would be useful for businesses that need strict control over the chatbot’s responses.

E.g. if someone asks if they can still add something to their order, they can before 2:00 pm, but after 2:00 pm they can't. It would be nice to have the AI Agent know what time it is and give an answer based on that.

Based on the Chat widget, the AI Agent can see on which page the website visitor is on, and use this information within the conversation.

We would like to have the option to choose a different LLM per AI Agent

For example, if a website visitor reopens the chat 24 hours later, the conversation from the day before is still in progress, which is not desirable in certain scenarios.

In the interactive tester you cannot add any attachment to test this behaviour. It would be good to also test this feature here.