Nous Research’s Hermes Agent earned its following in the terminal — but terminals scare off most people. On June 2, 2026, the project shipped Hermes Desktop, a native app for macOS, Windows, and Linux that puts the same self-improving agent behind a clean graphical interface. Crucially, it isn’t a stripped-down clone: it’s “another surface over one agent, not a fork.” Here’s exactly what Hermes Desktop is and who should use it.
Key takeaways
- Hermes Desktop is the official native app (macOS, Windows, Linux) for Nous Research’s Hermes Agent, shipped June 2, 2026 as a public preview (v0.15.2).
- Same core, new surface: it shares the identical agent — config, API keys, sessions, skills and memory — with the CLI. Start a chat on desktop, continue it in the terminal.
- Built for humans: streaming tool output, a side panel to preview web pages and files, voice input/output, drag-and-drop files, and an inline model picker.
- Provider-agnostic: connect local Ollama/vLLM, any major API, or Nous Portal’s bundled “300+ models.”
- Best for: anyone who wants an autonomous, memory-keeping agent without command-line setup.
What is Hermes Desktop?
Hermes Desktop is a graphical front end for the open-source Hermes Agent — the autonomous AI system from Nous Research that plans, acts, and observes in loops to complete tasks while keeping memory across sessions. Where the original Hermes lived in a CLI, the desktop app delivers that same agent as a real, installable application with a window, buttons, and a settings screen.
It launched as a public preview at version 0.15.2 on June 2, 2026, with native builds for all three major desktop platforms.
“Another surface over one agent, not a fork”
This is the most important thing to understand about Hermes Desktop, and the line Nous Research uses to describe it. The desktop app reuses the identical agent core as the CLI and every messaging integration. That means it shares the same:
- Configuration and API keys
- Sessions (conversations)
- Skills (the capabilities the agent has learned)
- Memory (everything it remembers about you and your work)
The practical payoff: you can start a conversation on the desktop and seamlessly continue it via the CLI or text UI, and vice versa. The desktop isn’t a separate, dumbed-down product with its own memory — it’s a window onto the same agent. Everything that makes Hermes Agent special — its persistent memory and self-improving learning loop — is fully present in the desktop app.
Key features
The desktop app is designed to make an autonomous agent legible to a human watching it work:
- Streaming responses and live tool activity — you see the agent’s output and its tool calls in real time, not after the fact. This “streaming tool output” was the headline feature of the v0.15.2 launch.
- A side panel for previewing web pages and files the agent is working with.
- Voice input and output — talk to the agent and have it talk back.
- File browsing and drag-and-drop — drop a file straight into the chat.
- An inline model picker in the status bar, so you can switch the underlying LLM mid-session.
- Concurrent multi-profile sessions — run more than one agent profile at once.
- In-app self-update and themes — it keeps itself current and lets you restyle the UI.
- Remote gateway connection — connect the desktop app to a Hermes agent running elsewhere (for example on a VPS) over OAuth or a username/password, so the heavy lifting can happen on a server while you drive it from your laptop.
Installing it across platforms
Hermes Desktop ships native builds for all three desktop operating systems, with slightly different install paths:
- macOS and Windows get direct installers — download, run, done.
- Linux uses a terminal-based installation for distributions.
Because it’s still a public preview, expect rough edges and frequent updates — the in-app self-updater exists precisely because the project is moving fast.
Connecting your models
Like the underlying agent, Hermes Desktop is provider-agnostic — it works with essentially any LLM backend. You have three broad options:
- Local models via Ollama or vLLM — the most private route, keeping everything on your own hardware. Pair it with one of the best local LLMs to run on Ollama for a fully offline agent.
- Your own API keys — OpenAI, Anthropic, OpenRouter, and others, for maximum capability.
- Nous Portal — Nous Research’s own subscription that bundles APIs across tiers (Free, Plus, Super, Ultra) with access to a reported 300+ models plus integrated tools like web search and image generation, so you don’t have to configure separate third-party APIs.
Hermes Desktop vs the Hermes CLI — which should you use?
| Dimension | Hermes Desktop | Hermes CLI |
|---|---|---|
| Setup difficulty | One-click install | Terminal / config files |
| Best for | Newcomers, daily use | Developers, automation |
| Visual tool activity | Streaming + side panel | Text output |
| Voice input/output | Yes | No |
| Agent core (memory/skills) | Identical | Identical |
| Scripting / headless server | Limited | Full |
The honest answer is that it’s not really either/or. Since both surfaces drive the same agent, many people use the desktop app for interactive, hands-on work and the CLI (or a VPS deployment) for automation and always-on tasks — switching between them freely because the memory follows.
Hermes Desktop vs other desktop AI apps
It’s easy to lump Hermes Desktop in with apps like LM Studio, Jan, or the ChatGPT desktop client, but they solve different problems:
- LM Studio and Jan are primarily chat front ends for running local models — excellent for talking to an LLM on your own hardware, but they are not autonomous agents with persistent memory and a self-improving skill loop.
- The ChatGPT or Claude desktop apps are polished cloud chat clients tied to a single provider, with no local-first option and limited tool autonomy.
- Hermes Desktop is a front end for an autonomous agent — it plans, calls tools, runs in loops, remembers across sessions, and works with any provider, local or cloud.
In short: if you just want to chat with a local model, LM Studio or Jan are simpler. If you want an agent that acts, remembers, and improves — and you’d rather not touch the terminal — Hermes Desktop is built for exactly that. You can even use Ollama or LM Studio as the model backend running behind Hermes.
Limitations to keep in mind
It’s a public preview, so expect bugs, breaking changes, and frequent updates (hence the in-app self-updater). As with any autonomous agent that can run tools, send messages, or execute code, scope its permissions carefully — give it only the access a task actually needs. And running a capable model locally behind it still demands real hardware; check our Ollama system requirements guide before committing to a fully local setup.
FAQ
What is Hermes Desktop?
Hermes Desktop is the official native desktop application for Nous Research’s Hermes Agent, released June 2, 2026 as a public preview (v0.15.2). It gives you the self-improving, memory-keeping Hermes Agent in a graphical app for macOS, Windows, and Linux — no command line required.
Is Hermes Desktop different from Hermes Agent?
No — it’s the same agent with a graphical interface. Hermes Desktop is “another surface over one agent, not a fork.” It shares the identical config, API keys, sessions, skills, and memory as the CLI, so you can start a task in the app and continue it in the terminal seamlessly.
What operating systems does Hermes Desktop support?
macOS, Windows, and Linux. macOS and Windows have direct installers; Linux uses a terminal-based install. It’s currently a public preview, so it updates frequently via an in-app self-updater.
Can Hermes Desktop run local models?
Yes. It’s provider-agnostic and connects to local models through Ollama and vLLM, keeping everything on your own hardware. You can also use hosted API keys (OpenAI, Anthropic, etc.) or Nous Portal’s bundled models.
Is Hermes Desktop free?
The app itself is part of the open-source Hermes project. You pay only for the LLM you connect — which is free if you run a local model via Ollama/vLLM. Nous Portal is an optional paid subscription (with a free tier) that bundles models and tools for convenience.
Should I use Hermes Desktop or the CLI?
Use the desktop app if you want the easiest setup, voice, and a visual view of what the agent is doing — ideal for newcomers and interactive work. Use the CLI for scripting, automation, and always-on server deployments. Because both drive the same agent and memory, you can use both and switch freely.
Bottom line
Hermes Desktop is how a respected but terminal-bound power tool reaches everyone else. By wrapping the Hermes Agent in a native app — with streaming tool output, voice, a side panel, a model picker, and the option to connect to a remote agent on a server — Nous Research kept everything that made Hermes special (persistent memory, self-improving skills, local-first privacy, 20+ providers) while removing the one barrier that kept most people out: the command line.
It’s still an early public preview, so expect it to change quickly. But if the idea of a private, self-improving AI agent appealed to you and the setup didn’t, Hermes Desktop is the easiest way in today.