compute.cx lets AI agents rent GPUs across Runpod, Vast.ai and Hot Aisle via CLI and MCP

hargup13 · x · 2026-11-08

Responding to gakonst's call for GPU providers willing to be paid by agents, the builder of compute.cx demonstrates a CLI + MCP layer for humans and agents to rent GPUs across Runpod (H100-SXM), Hot Aisle (MI300X 192GB), and Vast.ai spot (AWS Trainium and GCP TPUs coming). A run looks like compute run train.py::train --gpu H100: locked rate, confirmation, provisioning, execution, receipt, then the machine is destroyed. Billing is provider pass-through plus a small fee (sample H100 job: $0.86 total), with $0 for runs that never become ready. Ships templates for SFT fine-tuning, preference training, and batch inference; the author says his agents use it daily for compiler research.

Original post →

More from coding & agent

coding & agent channel →