So I'm working on a bot and I've realized that I have a lot of Kb's. My bot has been innacurate in the way it's been answering lately so could this be due to the fact I have so many Kbs?

Did you specify which KB the bot should get the answer from

I have it set so that it looks in every Kb

Hey, to get a better handle on what’s happening, could you share some examples of the questions and answers where the bot went off track? Also, knowing how many knowledge bases (KBs) you’re working with would be helpful. It's possible that narrowing down the search to specific KBs through the Knowledge Agent override on your workflow nodes might improve accuracy. Have you experimented with using GPT-4 in your KB agent configuration? Selecting "Best Model" might yield better results. Don’t forget to utilize the logs and the debugger panel too. By inspecting an answer your bot generated, you can see the search results it pulled from your KBs. Check if the relevant documents or chunks are being considered in the search results. This insight could be key to pinpointing where the inaccuracies are stemming from.

Here is an example in which the query responses are polluted by queries from other knowledge bases, located in other folders at the same hierarchical level, which should not be queried. In the image, the instructions for the query indicating the knowledge base to consult (“MODEL 5C RULES”). As you can see in the other image in the JS5REGOLE folder the knowledge base (“MODEL 5C RULES”). In the event debugger you can see the outcome of the queries which are actually made on all the kb, even those contained in other folders. And this pollutes the answers! Furthermore, searches are done in GPT-4. But this should be irrelevant because the correct functioning of the query should not be conditioned by the model…

Understood! The issue seems to stem from using the Query KB Agent card. As highlighted in this [post](https://discord.com/channels/1108396290624213082/1218301695831052441/1218301695831052441), the problem is due to outdated documentation related to folder structure. We’re sorry for the mix-up and will update the documentation soon. Also, we’re planning to enhance the Query KB Agent card by adding the ability to select specific KBs. Stay tuned! [https://discord.com/channels/1108396290624213082/1218301695831052441/1218301695831052441](https://discord.com/channels/1108396290624213082/1218301695831052441/1218301695831052441)

Thank you! I read from the post that the problem will be solved next week... in the meantime, to get correct results, what can I do?

# Sending a Problem Report Here's how to send the Botpress Team a problem report: 1. In the Botpress Cloud Studio, locate the "Help" icon in the bottom left corner of the screen. 2. Click "Report a problem" and follow the steps on the screen. 3. Copy the Report ID to your clipboard and provide it to a member of the Botpress Team on Discord.

If it makes sense in your flow, you could try and only select a few KBs. Just to clarify, the responses and response rate from the KB can only be as good as your KB is. I would recommend you to watch this video: