It barely needs 2gb, it a simple process that glues external APIs... a 4GB VPS would have plenty of spare headspace.

Running a local model via ollama etc is a different can of worms, need 16gb of VRAM or 128GB unified DDR5 to do that even half-well.

what makes it shitty tho?

At these price points your not getting a real SLA and probably hardware that's over-provisioned and insufficient redundancy for critical workloads... if something is really important you gotta go to an Akamai-tier vendor