Hermes Agent with Ollama: local setup and GPU guide

Facundo Corengia walks through installing Hermes Agent with Ollama and Docker for a self-hosted AI agent. The tutorial covers downloading the project, choosing a local LLM, configuring a custom endpoint, and starting a terminal chat. Messaging connections such as Telegram and WhatsApp are left for later videos.

The demonstration uses Windows. Corengia recommends 16 GB of RAM and at least 15 GB of free storage, while describing 8 GB as a constrained option. He also discusses Windows 10 or 11 and Macs with an M1 or newer chip. Initial installation requires internet access. The guide runs Ollama inside Docker on Windows and Linux, but uses native Ollama on Mac. Git is optional because viewers can download the project as a ZIP file.

Model choice and hardware affect the result. Corengia discusses Qwen 3 with 8 billion parameters for a 16 GB machine and selects Qwen 3.5 during configuration. His CPU attempt produces no response after five minutes. He then switches to Gemma 4 12B and enables NVIDIA GPU support through a Docker Compose override. Since both change, the demonstration does not isolate the GPU's effect on speed.

Corengia describes the setup as private, but the tutorial does not independently verify that claim. He also warns about Docker's memory use and advises trying a smaller model when performance is poor.