Hermes just got a lot more accessible for developers who prefer their tools in a browser. Nous Research has shipped an official Web UI for its self-hosted AI agent, giving users a full-featured alternative to the command-line interface without sacrificing any of the underlying capabilities that have earned the project 126k+ GitHub stars.

A Three-Panel Layout That Mirrors Claude Code

The new interface follows a familiar Claude-style layout with three distinct panels: a sessions sidebar for conversation history, a central chat window, and a workspace file browser. Unlike many web UIs that ship as stripped-down versions of their CLI counterparts, Hermes Web UI promises 1:1 parity with the terminal experience โ€” everything you can do from bash, you can do from Chrome. Key features include multi-profile support so multiple users can maintain separate identities on the same instance, session history organized by projects and tags, inline file preview for code, markdown, and images, light and dark mode themes, password protection for shared servers, voice input via the Web Speech API, and mobile-responsive design. The Docker image is available at ghcr.io/nesquena/hermes-webui:latest, making deployment straightforward for anyone already running Hermes in containers.

Why This Matters for Self-Hosted Workflows

The core value proposition of Hermes has always been persistent memory that compounds across sessions โ€” not just context within a single chat, but accumulated knowledge about your stack, conventions, and workflow that survives reboots and model swaps. That memory lives as editable markdown files in ~/.hermes/, which means you can inspect, modify, or nuke it at will. The Web UI doesn't change this architecture; it just makes the experience more approachable for users who want to check on running agents, browse session history, or manage workspace files without SSH-ing into a server. For teams sharing a single Hermes instance, the visual file browser and project tagging system add some welcome organization that the terminal interface lacks.

Comparison: Where It Stands Against the Competition

Nous Research doesn't shy away from direct comparisons in their documentation. The head-to-head analysis positions Hermes against OpenClaw (MIT, 365k+ stars), Claude Code, Cursor, Claude.ai/ChatGPT, and Perplexity Computer โ€” arguing that while each competitor excels at specific tasks, none of them offer the same combination of self-hosted execution, persistent cross-session memory, autonomous scheduling on your own hardware, multi-surface messaging reach, and automatic skill improvement from experience. The Web UI specifically targets users who want the power of a server-resident agent without living in tmux. It's a pragmatic move: not everyone wants to SSH just to ask their AI assistant a question, but they also don't want to hand that conversation to a cloud service where it lives forever on someone else's servers.

Getting Started

Installation requires cloning the hermes-webui repository from GitHub and running ./start.sh, or pulling the Docker image with the appropriate volume mounts for ~/.hermes/ and ~/workspace/. The Web UI runs on port 8787 by default and is SSH tunnel ready for secure remote access. Existing Hermes users just need to point their browser at the server running their agent โ€” no additional configuration required if the CLI is already operational.

Key Takeaways

  • Hermes Web UI delivers full CLI parity in a browser with sessions, file browsing, and voice input
  • The interface targets developers who want self-hosted AI without terminal dependency
  • Multi-profile support enables team sharing on a single persistent instance
  • Docker deployment makes it practical for users already running Hermes in containers

The Bottom Line

The Web UI is exactly what the ecosystem needed โ€” not because CLI is hard, but because different tasks call for different interfaces. A polished browser view doesn't undermine Hermes's self-hosted philosophy; it extends it to workflows where opening a terminal feels like overkill. If you've been on the fence about running your own agent instead of paying for Claude.ai or ChatGPT Plus, this lowers one more friction point.