Hello, Convai team.
We are facing a significant limitation regarding the use of the Knowledge Bank. We’ve been using the platform for over a year with several characters, and even after uploading, training, and connecting files via the Knowledge Bank, the characters are unable to accurately retrieve the information contained in those documents.
Even after connecting the files and confirming they are marked as “File available,” the character replies that it doesn’t know, wasn’t trained, or has no knowledge of the topics covered in the uploaded files. This forces us to insert all relevant content directly into the Backstory field, which seems to be the only input actually interpreted and used during interactions.
We would like to know:
1. Does the model used by Convai actually perform semantic search across documents connected through the Knowledge Bank?
2. Is there any additional configuration required to ensure that the file content is considered in the responses?
3. Are there any size, format, or structural limitations that prevent the model from accessing the content?
4. Is there any planned improvement or upcoming update to make this functionality work contextually, so characters can truly leverage the information in connected files?
Thank you in advance. We look forward to receiving a detailed technical response.
