Hey, @basecluesurf! I came across your thread and thought I’d have a go at answering 
The short answer: for the kind of architecture you’re running, your current hybrid setup is exactly right.
Vercel has made major strides toward handling backend workloads, but it’s not a full replacement for persistent servers (like DigitalOcean, AWS, or GCP). For use-cases such as:
- Long-running or heavy compute tasks: Anything exceeding Vercel’s max 13-minute runtime (e.g., hours-long video processing, large batch jobs).
- Persistent WebSockets: Vercel Functions can’t hold always-on connections, so real-time bidirectional servers must run elsewhere.
- Self-hosted services: Tools like Meilisearch, Grafana, Kafka, or Temporal require containers/VMs and cannot run on Vercel.
- Always-on workers: Background daemons (Kafka consumers, Temporal workers, etc.) need persistent compute.
- Complex media processing: Memory-intensive or multi-hour ffmpeg/Docker workflows are not supported in a serverless environment.
Appreciate the feedback, and of course, we’ll share this with the product teams as we continue to build. I’ll keep this thread updated as well. Hope that helps!