Hermes Agent + Ollama: local setup with Qwen

This tutorial walks through using Ollama to power Hermes Agent on a Mac. The speaker installs Ollama, downloads a Qwen model and tests a conversation before connecting a model to Hermes. The aim is to run models locally, with the speaker presenting privacy and offline use as reasons to choose this setup.

Model selection needs more attention than the initial download. The speaker asks Hermes for recommendations using a screenshot of the MacBook's specifications, then discovers that the first model does not meet Hermes's stated 64,000-token context requirement. He switches to Qwen 3 Coder 30B and shows a configuration labeled 64K in Hermes. The walkthrough does not establish a minimum RAM requirement.

The optional Hermes desktop app provides a session interface and chat branching. When it fails to launch during setup, the speaker updates Hermes through the terminal and retries successfully. He also mentions Telegram as another way to chat with Hermes.

The final discussion weighs a local LLM against cloud services. The speaker warns that larger models can slow the computer and recommends leaving hardware headroom. He proposes local processing for sensitive documents and cloud models for demanding tasks or fresh web information. His claims of complete privacy and zero token charges are part of the presentation, rather than an independent security or cost assessment.