Just kidding, of course! But it does seem like a feature that would be worth while to seriously consider. I can imagine this will take extra processing power on your end so to cap the refresh at say once a week as suggested above might be a good balance.
Hi team! What do you think about adding support for gzipped sitemaps in our model training pipeline? It’s a widely adopted standard. If you agree, please take a moment to vote on the idea.
Would be so helpful if the chatbot could update itself as it is hard to keep retraining the chatbot every time something new is added or updated on multiple websites.
We need to be able to connect the chatbot to product feed and data feed. So that all data is updated automatically, not only those listed in the web pages section limited to 200 web pages.
Automatic product and data feed is essential for feeding correct data to chatbot
Unfortunately, automact updating of the website URLs added to the knowledge base isn’t supported now. This idea is already on the Wishlist and I’ve added your comment there too.
As for now, the only option is to retrain each page manually:
I initially solved the problem with automatic updates by adding the following instruction to the conversation rules:
Before making any statements that require an up-to-date knowledge base (in my case, a web link with current JSON data for the bot), this data must be retrieved. The bot should only provide the requested information and not display the entire content of the link. Furthermore, the response must not be based on the previous chat history.
To test this functionality, I also added the instruction to inform the chat participant that “…the data is being retrieved…”
So far, this has been a successful test. However, during testing I noticed that the bot sometimes couldn’t check this. The reason isn’t clear yet. A simple trigger to start the tetraine process should still be achievable.
I’ve discussed your use case with the devs and they’ve said it must be a coincidence. Currently, it’s impossible to make the ai assistant retrain links using the instruction prompts. Creating a custom trigger for the automatic URL updates isn’t feasible for now either.
I hope this clears things up.
If you have any questions or additional feedback, we’re always here to discuss
I have events on my website that change weekly. I have noticed the chatbot still suggests a past event. What would I need to do to get the chatbot to check for the latest events each time it’s asked? I did make a rule for that, but it did not work.
Unfortunately, there is no way to automatically update the knowledge base after chages made on the website. However, we understand the importance of this feature. This idea is already on the Wishlist and I’ve merged your comment with it.
As for now, the only option is to manually retrain the pages where the changes were made:
Awesome, thank you. If either you or your developers can benefit from the specific technical implementation details (code, flows, testing, quality check procedures, etc.), please send me a DM. Our team can provide them.
This is not a biggie for us (our company), but certainly can be for administrators managing multiple sites at one time.
Imagine if you had 117 different websites under your responsibility using the Elfsight AI Chatbot. Now let’s do the math: Imagine clicking “Retrain” for each website that has 5 or 6 pages that are updated every week. That’s 117 x 5 = 585 clicks every week. Ouch!