Hey everybody, What is the best way to make sure the agent follows instructions ? Is it the main prompt ? Is it the knowledge badge ? For example, sometimes the agent doesn't fetch the knowledge badge and hallucinates instead. It responds with incorrect information, randomly. On other times, it gives a different reply. How to make it reliable and not reply random answers ?

To ensure your Vapi AI agent reliably follows instructions, use a clear and strict system prompt that directs it to rely only on the knowledge base, ensure the knowledge files are correctly uploaded and indexed, test frequently, monitor call logs, and refine both prompts and content regularly to prevent hallucinations or inconsistent replies.