Researchers have released a new study Hermes Desktop in public preview. It’s a native app for macOS Windows Linux. The open-source Hermes Agent now has a graphical user interface. Hermes was previously run through CLIs and message gateways. Hermes Agent is currently v0.15.2.
The desktop, according to Nous Research documentation reuses the agent core. It is able to share memory, configuration, session, skill, API key, and other information with CLI. It is not a separate agent but merely another surface.
Hermes Desktop
Hermes agent is an independent AI agent. This is not an editing copilot. It executes tasks, invokes tools, and retains the state of a session across sessions. Here, an agent is a model which plans, executes, and observes continuously.
Hermes Desktop, a GUI built on the same core agent software. No terminal is required to run it. It shows live response and tool activity. The right-hand panel shows previews of web pages, documents, and output from tools. The UI also features a file manager, voice input/output, and settings.
Session sharing is possible across all surfaces. The CLI/TUI will continue a conversation that was started on your desktop. Because state isn’t duplicated, the reverse works as well.
MacOS and Windows both offer direct installers. Linux installations can be done from the Terminal on any distribution. A script that installs Linux. --include-desktop Build the app using an existing installation.
Closed Learning Loop
Hermes is described by our research team as having an closed learning loop. That’s what makes it different from a chat wrapper. The agent creates reusable skills after completing a complicated task. The skills improve over time.
Memory persists and is curated by agents, who send periodic reminders for knowledge saving. FTS5 search sessions are combined with LLM summary for cross-session memory. Honcho dialectic User Modeling is used for user modeling. Longer usage means more context is retained and reused. Agentskills.io is an open standard for skills.
What It Does, Sandboxes and Schedules
Hermes allows you to manage multiple messaging systems from a single gateway. Telegram is listed in the desktop. It also includes Discord, Slack and WhatsApp. Start a task in one platform, and then continue it on another.
Schedules are written in natural language. This is used for backups, reports and briefings. They are run by a built in cron scheduler through the Gateway.
Subagents are created by delegation. They have their own terminals and conversations. Subagents are separate workers who handle a single task. Python RPCs reduce multi-step pipes to a few zero-cost, context-free turns.
The execution is sandboxed. There are five options for the backend: SSH (local), Singularity (SSH), local and Docker. The container is hardened and the namespace isolated. Namespace separation limits the information that a process running can access or see.
Included in the software are web search, automation of browsers, images, speech-to-text, multi-model reasoning, and vision. Hermes can also be connected to external tools via MCP. MCP stands for Model Context Protocol. It is a standard that allows tool integration.
API Keys are not required for Hermes. Portal instead bundles these services into one subscription. There are four portal tiers – Free, Super, plus, and ultra. Monthly credits are included in the paid levels, as is access to more than 300 models. These tiers also come with built-in tools.
Tool Gateway is a tool that allows multiple tools to be accessed from one account. Firecrawl is used for web searches, and FAL for image creation. OpenAI uses Text-tospeech, and Browser Use is the cloud browser.
Strengths, Questions and Answers
Strengths:
- For most users, native installations eliminate the need for a terminal.
- The tool call previews and streaming output make it easier to see the results.
- The ability to improve and maintain a persistent memory reduces the need for repeated instructions
- The design is not model specific, so you aren’t locked in to one provider.
- The MIT License allows for auditing, self-hosting and modification
Questions:
- This product is public preview so there will be some rough edges
- Questions about oversight and review arise from the autonomous memory and scheduling.
- Linux Desktop still requires the Terminal to install
- Beginners will have a more steep learning curve because of the wide range of capabilities
The Key Takeaways
- Nous Research launched Hermes Desktop as a public preview. It is a macOS native app, Windows desktop and Linux application for Hermes Agent, the open source software.
- With the CLI, the GUI and Gateway share a single agent core and configuration. Sessions, skills and memory are shared across all surfaces.
- The no-terminal version includes a preview pane on the side, a file browser, audio I/O and settings.
- Hermes can be used with any endpoint, including OpenRouter and OpenAI.
- Hermes Agent is currently v0.15.2. It comes with a closed-loop learning system, MCP support and 5 sandboxes.
Check out the Project here. Also, feel free to follow us on Twitter Don’t forget about our 150k+ ML SubReddit Subscribe Now our Newsletter. Wait! What? now you can join us on telegram as well.
You can partner with us to promote your GitHub Repository OR Hugging Page OR New Product Launch OR Webinar, etc.? Connect with us

