What Hermes Agent Desktop can do, how to pick a provider or a local model through LM Studio, where data is stored, and how the app differs from Doka.

Hermes Agent was long associated with the terminal, but the project now has a separate desktop app for Windows, macOS, and Linux. It runs the same agent, shows files and tool results in an interface, and lets you pick a model on first launch. Here's exactly what stays local, and who this format fits.

What Hermes Agent Desktop is

Hermes Agent is an open agent from NousResearch. It works with files and the terminal, uses skills, memory, scheduled tasks, and pluggable tools.

Desktop isn't a simplified chat bolted onto another product. Per the official app documentation, it uses the same agent, configuration, sessions, skills, and memory as the CLI and gateway. A local hermes serve backend runs inside, so no separate terminal is needed for everyday use.

What's in the app

The desktop interface lets you:

browse the files in your working folder;
see code output, tables, charts, and other previews;
switch profiles and models;
manage skills and scheduled tasks;
continue the same sessions used in other Hermes interfaces;
connect to a remote Hermes backend instead of a local one.

That last point matters: the app window can be on your laptop while the agent itself runs on a server. In that mode, only the interface is local — not the computation, and not necessarily the data.

How to install Hermes Agent Desktop

Only download the installer from the official releases page. The project publishes builds for Windows, macOS, and Linux; available formats depend on the specific release.

On first launch, the app walks you through onboarding: pick a provider and model, check the working folder and permissions. Don't jump straight into an important repository — give it a test folder first and see which commands the agent proposes running.

How to connect a local model

Installing Hermes locally doesn't mean a local model. Pick OpenRouter, Anthropic, Gemini, or another cloud API, and requests and the context you pass along will go to that provider.

The official provider list includes ready-made integration with LM Studio. Start a local server in it, load a model, then pick LM Studio during Hermes's setup. How LM Studio differs from Ollama in this role is covered in a separate comparison. For another OpenAI-compatible local server, you can use a custom provider; check the address and exact parameters against the current provider documentation.

◆

Ollama Cloud and local Ollama aren't the same thing. The cloud provider sends requests to an external server. For a private setup, check that the endpoint points at localhost or a machine you control.

Which model to pick

Hermes uses tools and runs long sessions, so the model needs enough context and reliable tool calling. A small model might answer well in chat but mix up command arguments or lose the original goal after a few steps.

Start with a short task with a checkable result. If the model messes up specifically at tool calls, bump up its size or simplify the set of available actions. For coding tasks, there's a breakdown of local Qwen Coder, and the GGUF file format is explained in a separate article.

Privacy and permissions

Sessions, memory, and the local backend's config live on your computer. But every cloud provider, remote backend, and external tool creates its own data channel. GitHub, Notion, or email don't become local just because Hermes runs on a laptop.

The agent can also execute commands. Pick a working folder explicitly, review diffs, and don't run the app with admin rights unless you need to. Keep a human confirmation step for irreversible actions.

Hermes Agent Desktop or Doka

Hermes suits people who want one open agent across several shells: desktop, CLI, gateway, and voice. It's flexible, but setting up a provider and local server stays part of the workflow.

Doka focuses on a simple desktop start. A local GGUF model can be picked and downloaded right inside the app, and the agent works with code, documents, spreadsheets, and the terminal. If you need a CLI or a shared backend for several clients, Hermes is closer to the job. If you want to install an app and start working in one folder right away, Doka gets you to a first result faster.

How to compare without a long setup

Take the same folder and one task: read three files, put together a summary, and save it alongside them. Use comparable models, block outside network access, and compare where it's clearer to pick files, confirm actions, and check the result.

Download Doka for free. Other open options for comparison are covered in the roundup of Goose and terminal agents.

Hermes Agent Desktop: Setup and Local Models