For makers and artists building autonomous workflows, the latest update from Nous Research is not just a new interface, but a fundamental shift in how you interact with AI agents. They have launched Hermes Desktop in public preview, a native application for macOS, Windows, and Linux that finally gives the open-source Hermes Agent a proper graphical interface. Previously, users were forced to run the agent via command line or messaging gateways. This new build, based on Hermes Agent v0.15.2, brings the tooling directly to your machine.
The Core Concept
According to the documentation, the desktop application does not fork the technology; it simply adds a new surface to the existing agent core. It shares everything—configuration, API keys, sessions, skills, and memory—with the CLI and gateway versions. This means the agent remains a single, unified entity regardless of the interface you choose.
Hermes Agent itself is designed as an autonomous actor rather than a simple coding copilot. It plans, acts, and observes in a continuous loop, executing tasks and calling tools while maintaining state across sessions. The new desktop GUI removes the need for a terminal. The main window displays streaming responses and live tool activity, while a right-hand pane previews web pages, files, and tool outputs. The interface also features a file browser, voice input and output, and a dedicated settings UI.
Crucially, sessions are shared across all surfaces. A conversation started in the desktop can be resumed in the CLI or TUI, and vice versa, because the state is never duplicated.
Installation is straightforward: macOS and Windows users get direct installers, while Linux users install via terminal on any distribution. An install script with an --include-desktop flag allows you to build the app against an existing installation.
The Closed Learning Loop
Nous Research distinguishes Hermes from simple chat wrappers by emphasising its closed learning loop. After completing a complex task, the agent writes a reusable skill. These skills then self-improve during subsequent use.
Memory is persistent and agent-curated, with periodic nudges to save knowledge. Cross-session recall utilises FTS5 session search combined with LLM summarisation. User modelling runs through Honcho dialectic user modelling. In practice, the longer you use the system, the more context is retained and skills are reused. These skills adhere to the agentskills.io open standard.
Connecting, Scheduling, and Sandboxing
Hermes operates across multiple messaging platforms from a single gateway. The desktop lists Telegram, Discord, Slack, WhatsApp, Signal, Email, and CLI. You can initiate a task on one platform and continue it on another.
Scheduling is handled through natural language instructions for reports, backups, and briefings. These run unattended through the gateway using a built-in cron scheduler.
Delegation spawns isolated subagents with their own conversations and terminals. A subagent acts as a separate worker handling a single job. Python RPC scripts collapse multi-step pipelines into zero-context-cost turns.
Execution is strictly sandboxed. The desktop lists five backends: local, Docker, SSH, Singularity, and Modal. It applies container hardening and namespace isolation, which limits what a running process can see or touch.
Built-in tools include web search, browser automation, vision, image generation, text-to-speech, and multi-model reasoning. Hermes also connects external tools through MCP (Model Context Protocol), a standard for tool integration.
Nous Portal and the Tool Gateway
Hermes works with any provider, making API keys optional. However, Nous Portal bundles them under one subscription. Portal tiers are Free, Plus, Super, and Ultra. Paid tiers include monthly credits and access to 300+ models, plus built-in tool use.
The Tool Gateway routes several tools through one account. Web search uses Firecrawl, image generation uses FAL, text-to-speech uses OpenAI, and the cloud browser uses Browser Use.
The next evolution of Hermes Agent is here!
Introducing Hermes Desktop: everything you love about Hermes, now native on your machine.
First demoed in Jensen’s GTC keynote, it’s now in public preview.
Strengths and Questions
Strengths
- Native installers remove the terminal requirement for most users
- Streaming output and previews make tool calls easier to inspect
- Persistent memory and self-improving skills reduce repeated instructions
- Model-agnostic design avoids lock-in to a single provider
- The MIT license allows audit, self-hosting, and modification
Questions
- The product is in public preview, so expect rough edges
- Autonomous memory and scheduling raise oversight and review questions
- The Linux desktop still installs through the terminal
- Broad capability means a steeper learning curve for beginners
Key takeaways
- Nous Research has released Hermes Desktop in public preview, a native application for macOS, Windows, and Linux that provides a graphical interface for the open-source Hermes Agent.
- The GUI shares a single agent core, configuration, API keys, sessions, skills, and memory with the CLI and gateway, allowing sessions to resume seamlessly across surfaces.
- The application requires no terminal, featuring streaming tool output, a side-by-side preview pane, file browser, voice I/O, and a settings UI.
- Hermes is model-agnostic and MIT-licensed, compatible with Nous Portal, OpenRouter, OpenAI, or any compatible endpoint.
- The current build is Hermes Agent v0.15.2, supported by a closed learning loop, MCP tool support, and five sandboxed execution backends.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
