Hermes Desktop and Ollama: Windows setup and comparison

This tutorial sets up Hermes Desktop with Ollama on Windows, then tests Gemma 4 against cloud Cowork on the same proposal-writing task. The speaker frames the local AI setup as an option for work that needs continuity or must keep client data on the premises.

The walkthrough installs both applications and starts a chat in Ollama to trigger the Gemma 4 download. It sets the context length to 64K because, according to the speaker, Hermes rejects a smaller context. Hermes then connects through a custom endpoint using Ollama's OpenAI-compatible API. The transcript does not supply the endpoint URL.

Both applications receive notes from a hypothetical agency discovery call. Gemma 4 produces proposal text in chat, but the demonstrated setup struggles to create a DOCX file through its tools. Cowork produces a formatted document. The speaker suspects further configuration could enable local file generation, but does not demonstrate a fix. GPU monitoring shows 10.3 GB of 12 GB dedicated memory in use while recording software also runs; this is an observation from that session, not a minimum requirement.

The closing discussion weighs hardware, maintenance and administration costs against control. The speaker recommends managed tools for most work and a local AI agent for narrower needs. Hermes' machine access and code execution still require guardrails; running a model locally does not establish that the surrounding workflow is secure.