Hermes Desktop and Ollama: setup, skills and local models

This tutorial covers Hermes Desktop installation, model connections and recurring workflows. The presenter describes Hermes as an open source AI agent with local memory files and reusable skills. The desktop interface handles the initial installation, while connecting Ollama later requires terminal commands.

The first setup uses a ChatGPT subscription through OpenAI OAuth. The presenter points to Codex device code authorization in ChatGPT's security settings as a possible fix for login errors. Using GPT 5.5, Hermes reads a sales spreadsheet from a local folder and produces a dashboard and PDF recommendations. It also creates a skill for repeating the analysis. A quarterly schedule reuses that skill, though the presenter warns that the computer must be active for scheduled work.

The final section explains how to run models locally with Ollama and select them in Hermes. The example machine is an Apple M1 MacBook Pro with 32 GB RAM. Model selection considers memory and 4-bit quantization. The presenter demonstrates an offline email rewrite with a quantized Qwen 3.6 model and the Humanizer skill, but questions how much text it removed.

The limits matter: the presenter reports that a local model failed to complete the earlier Excel workflow despite having its skill. Running the desktop app on your computer also does not make a cloud model connection private; the tutorial switches the model backend for local processing.