We’ve updated the AI Chatbot settings to make them clearer, easier to navigate, and more convenient to use!
Now, everything is better organized, so setting up your chatbot should feel much smoother. We really hope you’ll enjoy the changes!
What’s inside the update?
Agent tab
We’ve added a new Agent tab, where you can manage your chatbot’s behavior in one place.
Here, you can choose the agent’s role and tone, set the Response Length, and add Conversation Rules (alternative to the previous Agent Instructions section)
Quick Replies are now called Suggested Questions. They’ve also been moved to the Agent tab, together with the Greeting Message section:
[!note]
As part of this update, we’ve separated the agent’s role and tone from the Agent Instructions. The remaining instructions were moved into one Conversation Rules field.
Knowledge tab
Training section now has its own dedicated Knowledge tab.
This tab includes all knowledge sources: Web Pages, Files, Questions and Answers, Text Blocks, and a new section called Business Information.
I was hoping your developers would incorporate the much-needed improvement that’s been in your books for a year now: The CSS and JS fields are extremely difficult to navigate through and edit. They are extremely narrow and cannot be expanded horizontally. As they stand now, errors can be easily introduced into the chatbot. Can you guys please take care of that within a decent timeline?
For chatbot training purposes, it would be extremely helpful if Team Elfsight can place the LLM and version number used by Elfsight inside the AI Agent Profile field. For example: “Language Model: OpenAI GPT-5.4 mini”
Your new settings now has a section called “CONVERSATION RULES”. In the previous version, it simply said “TRAINING INSTRUCTIONS”, which is were I placed all of our rules (under one single stream of instructions). How does your new section affect bot training and the accuracy of its answers? Do I now have to list each rule separately? See below.
As always, a huge thank you for your feedback and continuous support!
I totally get your point regarding the enhanced navigation in the Custom CSS and JS editors and we also hope the devs will pay attention to this request in one of the future updates.
Thank you for the suggestion! I’ve added it to the Wishlist on your behalf and if the idea becomes popular, we’ll try to think it over - Display LLM version in the configurator.
Good question! Please let me consult with the devs and I’ll update you once I have their response
Thanks! Concerning Item 2., the developers should be able to add that information to the configurator pretty quick. Honestly, I would not consider it a feature request. The added field does not directly impact users since it’s not variable that users can control. Cheers!
The thing is that we have a set roadmap for the features we’re planning to implement next, and it’s not that easy to squeeze new ideas into it, no matter how good they are.
You’re right, it shouldn’t be too difficult to add it. However, since we consider both the popularity of the ideas and their technical difficulty, I’ve added this suggestion to the Wishlist.
If more users support this idea, we might prioritize it sooner, but we can’t give any promises for now.
Answering your third question, you can keep everything in a single rule, especially if you’re editing prompts in an external file or source and then copying them into the widget.
However, if you’re editing the instruction directly in the widget, organizing rules or scenarios by specific topics can help the agent follow the instructions more clearly.
Hi, Max!
First, I’d like to thank you all for the recent update. It’s clear this was designed to improve structure and long-term usability; however, I would like to share some important feedback based on my experience.
The update appears to have removed or reset previously entered Agent Instructions without warning. In my case, this included a fully developed chatbot system that had been carefully structured over time, including behavior rules, tone control, and guided response logic. Fortunately, I had a separate backup, but without it, this work would have been lost.
For users who rely on your platform to represent their brand professionally, this creates several concerns:
Loss of time and effort invested in building structured systems
Risk of unintentionally degraded chatbot performance
Lack of confidence when future updates may affect existing configurations
I believe this could be improved with a few safeguards:
Advance notice before updates that may affect stored configurations
Automatic backups or version history for instructions and rules
A migration preview or confirmation step before changes are applied
Clear documentation explaining how existing setups will be transitioned
The new structure itself is promising, but preserving user work and providing continuity is critical as well, especially for those using the chatbot as a core part of their business experience.
As always, thank you for your time and for continuing to improve the platform.
Thanks so much for sharing your thoughts with us – we really appreciate it!
I completely understand your concern, and I’m sorry for any inconvenience this has caused. Your suggestion for a migration preview is great, and we’ll definitely keep it in mind. Going forward, we’ll try to notify everyone before making any major changes.
As for the option to download the previous version of the instructions, I’m sorry it’s not an option now. The good news, however, is that it’s backed up on our end, and our devs can provide it if needed
Just to clarify how the migration was handled: we’ve separated the agent’s role and tone from the Agent Instructions, while the rest of the instructions were consolidated into one Conversation Rule.
Is it correct that everything worked like described on your end, but you were looking for the previous, full version of the instructions for other purposes?
Or were some important parts of the instruction missing from the Conversation Rules, and did you have to manually restore them from your saved version?
Thank you for the clarification and taking the time to explain how the migration was handled. I genuinely appreciate the transparency, as well as the reassurance that previous versions can be retrieved if needed.
If I had not maintained a backup of my original setup, I would have needed to request the previous version of the instructions to restore what was entered. Some parts of the original agent instructions were not carried over into the Conversation Rules, and I did need to manually restore them using my saved document. Because of this data loss, there was a noticeable difference in how the chatbot responded until those elements were reintroduced.
I appreciate the openness around both the update and its current limitations. It is through exchanges like this that collaborative efforts can continue to refine and improve the system as a whole, and I am glad to contribute where helpful.
It looks like something went wrong during the migration of your widget, since everything should’ve been moved to the Conversation Rule. I am genuinely sorry for this issue!
Just to confirm, is the bot working as it should after adding the missing parts?
