Even if VAPI provides multimodal LLM support to feed direct speech directly into language models, I'm concerned that it might actually make things worse rather than better. Introducing VAPI as an intermediary adds extra steps that may not be necessary if we can interact directly with the multimodal LLMs. Here are some downsides to consider: - **Additional Network Hop**: Establishing a WebSocket connection to the VAPI server, which then relays the speech to the multimodal LLM, adds an extra network hop. - **Increased Complexity**: Adding an orchestration layer introduces more complexity into the system without clear benefits. - **Higher Latency**: The extra hop could increase latency, affecting real-time performance. - **Extra Costs**: Using VAPI as an intermediary may lead to higher costs due to additional service fees and resource usage. - **Redundancy**: If we can feed direct speech directly into multimodal LLMs, using VAPI might be redundant. - **Potential Bottlenecks**: Introducing another layer could become a bottleneck or single point of failure. From a product development perspective on top of VAPI, these points are concerning. I'm trying to assess whether it's worth investing effort into integrating with VAPI given these developments. Is there something I'm missing about the advantages of using VAPI in this context?