Hey folks - I'm attempting to build a customer support agent. There are 3 primary parts to the call: 1. Identify the customer's issues 2. Identify which file / reference guide from the knowledge base is most relevant (pick only one) 3. Read the troubleshooting steps one-by-one I have #1-2 working, but when it gets to #3, the voice agent isn't following the troubleshooting steps from the specific reference guide in its knowledge base. Any guidance? Thank you in advance🙇🏻

Hi there! 👋 I understand you're having trouble with your customer support agent following troubleshooting steps from your knowledge base. I can't see your specific implementation details, but I can definitely help with common issues and solutions for this exact problem based on information I've found. Here are some potential solutions to explore: ### 1. Knowledge Base Configuration Make sure your knowledge base is properly configured and linked to your assistant: - Verify that files are correctly linked to your assistant's knowledge base settings - Check that file IDs are correctly referenced in your configuration - Ensure the knowledge base is active and properly connected to your AI assistant According to [troubleshooting information](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1336785070408732764), incorrect file linking is a common issue. ### 2. Prompt Engineering Your prompt structure might need adjustment to explicitly instruct the agent to read steps sequentially: - Break down the task into a clear sequence of steps - Explicitly indicate when the agent should wait for user responses - Include clear instructions to read troubleshooting steps one by one The [Vapi prompting guide](https://docs.vapi.ai/prompting-guide) recommends: "Breaking down the task into a sequence of steps enhances the agent's understanding and ensures a structured conversation flow." ### 3. Content Chunking Consider how your knowledge base content is structured: - Break down information into smaller, digestible pieces - Adjust the

targetSplitsPerChunk

to improve relevance of retrieved chunks According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1336243643043938395), content chunking can significantly improve how well agents follow step-by-step instructions. ### 4. Explicit Instructions in System Prompt Modify your system prompt to be very explicit:

Hey! 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 us figure out what went wrong!