Kaggle, Ollama and Hermes Agent setup on cloud GPUs

This tutorial runs Ollama on a Kaggle notebook and connects it to Hermes Agent on the user's computer. The model runs on cloud hardware, so this setup does not provide offline inference on the user's own machine.

The speaker describes free access to two T4 GPUs with 16 GB of memory each and a weekly allowance of 30 hours. Those are the video's stated terms, rather than a guarantee of current availability. Setup begins with account registration and phone verification, which the speaker says is needed for GPU access. The notebook's accelerator setting selects the two T4s.

The walkthrough installs Ollama in code cells, starts it with Ollama serve, and downloads a model through Hugging Face's Ollama option. The speaker reports trying Q4 and Q6 variants successfully, but the transcript does not identify the model clearly enough to name it.

A Cloudflare tunnel exposes the notebook's Ollama service through a public URL. In Hermes Agent, the speaker selects a custom endpoint, appends /v1 to that URL, and enters ollama as the API key value. A short chat demonstrates the connection. The speaker warns that an initial request may time out and suggests retrying, although that timeout does not occur in the demonstration. The closing claim of unlimited use conflicts with the weekly allowance stated earlier.