I'm unable to retrieve information from the knowledge base after updating it, even though it's in a valid Markdown format. The only workaround I've found is to use a specific instruction format like this: [Response Guideline] ... - To respond about the Restaurant Name, use Section 1 and the subsection "Name". [Here, I write the correct response format so the agent doesn't deviate from the script]." Is there a more efficient way to make this search work? We're spending too many hours just trying to get accurate results.

@Jefferson Petilo I'm sorry to hear that you're experiencing difficulties with retrieving information from the updated knowledge base. It sounds like you've already found a workaround by specifying instruction formats, which is great. However, to improve efficiency, you might consider the following strategies: 1. **Indexing and Categorization**: Ensure that the knowledge base is well-indexed and categorized. This allows for quicker search and retrieval of information. 2. **Search Functionality**: Implement a robust search functionality that can understand context and synonyms, which can help in retrieving the right information without needing to follow a strict instruction format. 3. **Regular Updates and Audits**: Periodically review and update the knowledge base to ensure all information is current and accurate. This can help reduce the time spent on searching for the right information. 4. **User Training**: Train users on how to effectively search and navigate the knowledge base. This can include using the right keywords, understanding the structure of the documents, and knowing how to interpret the results. 5. **Feedback Mechanism**: Implement a feedback mechanism where users can report any issues they encounter with the knowledge base. This can help identify and rectify problems quickly. 6. **Automation Tools**: Consider using automation tools that can help in organizing and retrieving information from the knowledge base more efficiently. If you've already tried these strategies and are still facing issues, it might be helpful to look into the specific search algorithms or retrieval methods you're using. Sometimes, tweaking these can lead to significant improvements in efficiency. If you need further assistance with the technical aspects of the knowledge base or its integration with your system, please ask support for more info on this.

1. You created a file or KB 2. You have to put something in there to refer to the KB regardless.

I uploaded a file and linked it to my assistant. The file is in markdown format (sample below). The assistant doesn't recognize the contents of the file when I update and re-link it to the agent. I can't share the exact contents because I'm under NDA. But the structure is like this: --- title: "Knowledge Base - X Restaurant" description: "Content optimized for VAPI" version: 1.0 date: 2025-01-04 08:00 AM --- # 1. GENERAL INFORMATION ## 1.1 Name - Restaurant Name ## 1.2 Mission - no no no no no no no no no no ## 1.4 About the Restaurant - Loren ipsum Loren ipsum Loren ipsum Loren ipsum ## 1.5 History - The Generic Restaurant Loren ipsum. --- # 2. LOCATION AND CONTACT ## 2.1 Address - 123 Main Street, Downtown, Cityville, CA ## 2.2 Phone - (123) 456-7890 ## 2.3 Parking - Private parking with valet service. --- # 3. OPERATING HOURS ## 3.1 General - Tuesday and Wednesday: 12 PM to 11 PM - Thursday, Friday, and Saturday: 12 PM to 12 AM - Sunday: 12 PM to 5 PM - Monday: Closed --- # 4. MAIN MENU ## 4.1 Appetizers ### 4.1.1 DISH 1 - **Ingredients**: ingredient 1, ingredient 2. - **Price**: $20 - **Allergens**: Alergen 1. ### 4.1.2 DISH 2 - **Ingredients**: ingredient 3, ingredient 3. - **Price**: $15 - **Allergens**: Alergen 2. --- The only way I could make it work was to put the all KB file content in the system prompt and put examples (and variations of answers simulating a real query) for ALL records in the knowledge base. This greatly increases our use of input tokens and can increase our cost. We spent many hours trying to make this simple feature work as expected. This is not the model we consider good.

Hey man I can help you with this if your still having trouble

Hey @Jefferson Petilo To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help me figure out what went wrong!