你安裝了 Hermes。現在讓它看起來比 ChatGPT 或 Claude 更出色

📄完整原文· 由 trafilatura 自動擷取10762 字

In brief - Nous Research's Hermes Agent crossed 100,000 GitHub stars in 10 weeks, spawning a fast-growing ecosystem of community-built GUI wrappers. - Four community interfaces now let users skip the command line entirely, from a Mac-native SSH companion to a full mobile PWA accessible via Tailscale. - All four GUIs run on a standard Hermes install and none require patching or forking the agent itself. So you installed Hermes. You ran it, asked it some things, it remembered them, maybe even built a skill on its own. Pretty good. But now you're staring at a terminal window and wondering if this is really it. It doesn't have to be. The Hermes community has been building user-interface wrappers at a pace that would embarrass most funded startups. Some of them are genuinely great. A few are beautiful. One of them might make your friends think you built something expensive. Here are the four best ones right now—what they do, what makes each one different, and how to actually get them running. #1 Hermes Desktop by Dodo Reach Repo: github.com/dodo-reach/hermes-desktop This one feels less like a GUI and more like a companion that just happens to live on your Mac. It's not trying to be a chat app. It's trying to be the place where you actually manage your agent. The whole thing works over SSH (secure shell, basically a way to connect to your computer remotely)—the same way Hermes already does. No gateway layer, no invisible sync process slowly drifting from what's actually on your server. When you open a session, skills tab, or cron job view, you're looking at live data from the real host. Nothing in between. What you get is a full overview of your active Hermes profile, session history, token usage, your skills library, your cron jobs, and a built-in terminal with multiple tabs. You can edit USER.md, MEMORY.md, and SOUL.md directly from the app—with a remote conflict check before it saves. You can run multiple agents on the same host side by side without losing track of which is which. The design is clearly written by someone who uses macOS every day. It feels native, behaves native. Version 0.5.0 added first-class cron job management and host-wide usage totals across profiles. The honest caveat: There is no chat interface. This is a management and monitoring app. You'll still type your messages in the terminal it gives you. For some people that's completely fine—it's a real shell, and the surrounding tabs give you context you'd never have otherwise. For others, it'll feel like half an app. One more thing: The app is not notarized by Apple yet, so macOS will warn you on first launch. Right-click → Open gets you past it. If Apple gets stubborn, go to your configuration settings and allow it to run under the security settings. It will appear as blocked. How to install: Download the universal build (Apple Silicon + Intel) from the Releases page. Unzip, move to Applications, right-click → Open on first launch to clear the Gatekeeper warning. You need SSH access to the machine running Hermes. That's it. #2 Hermes Desktop by Fathah Repo: github.com/fathah/hermes-desktop Same name, completely different project, completely different philosophy. This one is about getting you from zero to chatting as fast as possible. Where Dodo Reach's version assumes you already have Hermes running somewhere, Fathah's version does the whole thing for you. It runs the official Hermes install script, handles provider setup, and puts a working chat interface in front of you in a single flow. Double-click the app, follow the prompts, start talking. The feature list is extensive. Streaming chat with tool progress indicators, token tracking per conversation, 22 slash commands, session search, profile switching, a persona editor for SOUL.md, a cron job builder, and support for 16 messaging gateways—Telegram, Discord, Slack, WhatsApp, Signal, Email, and more. Model support covers OpenRouter, Anthropic, OpenAI, Google Gemini, xAI Grok, Nous Portal, Qwen, MiniMax, Hugging Face, Groq, and any local endpoint running LM Studio, Ollama, or llama.cpp. You can switch models from the UI, no config file editing required. The design is more generic than Dodo Reach's—less "native Mac," more "cross-platform product." But that cross-platform part is the actual point. Builds exist for macOS, Windows, and Linux. All three install the same way, all three give you the same feature set. If you want Hermes to look and feel like a proper chat app—something you could hand to a friend who has never opened a terminal—this is the one. How to install: Go to the GitHub repo, click Releases, download the build for your OS. On Windows, SmartScreen will flag it as unsigned—click "More info" then "Run anyway." On Fedora Linux, append --nogpgcheck to the install command if your system enforces GPG signature checking. Mac install is standard. #3 Hermes WebUI by Nesquena Repo: github.com/nesquena/hermes-webui If Claude's interface had a well-maintained open-source twin that ran entirely on your own server, it would look a lot like this. Hermes WebUI is a browser-based interface built with Python and vanilla JavaScript. No build step, no framework, no bundler. You run one command, it starts a local server, you open it in your browser. The whole thing is intentionally simple to deploy—and the result is an interface that feels complete without feeling heavy. The layout is three panels: sessions and navigation on the left, chat in the center, workspace file browser on the right. Model selection, profile switching, and workspace controls live in the composer footer as pill buttons—always visible, never buried in a menu. A circular token ring shows your context usage at a glance, with cost estimates on hover. You can pick your model from the UI, change the thinking speed (fast, extended reasoning, or automatic—essentially ChatGPT's fast vs. thinking modes but on your own infrastructure), and browse your agent's memory, skills, and session history without touching the command line. Mermaid diagrams render inline. Claude's extended thinking shows up as collapsible gold-themed cards. Seven themes ship out of the box: dark, light, dim, solarized, monokai, and OLED (pure black for burn-in prevention). Switch with /theme dark in the composer or through the settings panel. Custom themes are pure CSS—no Python changes needed. The project has 66 contributors and ships at a version-per-day pace. Recent additions include an embedded terminal, MCP server management UI, JSON and diff viewers, and live reasoning cards. How to install: The instructions are in the Github repository, but in general terms just clone the repo and run the bootstrap script. It detects whether Hermes is installed and, if not, runs the official installer for you automatically. git clone https://github.com/nesquena/hermes-webui.git hermes-webuicd hermes-webui python3 bootstrap.py Then open http://localhost:8080 in your browser. For remote or mobile access, set up an SSH tunnel from your local machine to the server. One command on each end and you're in from anywhere. #4 Hermes Workspace by Outsourc-e Repo: github.com/outsourc-e/hermes-workspace This is the most ambitious one. It's also the one that makes Hermes look better than most commercial products. It’s also the one we prefer, by a wide margin. The UI is clean, elegant, customizable, and powerful. Hermes Workspace was built during the Nous Hackathon 2026 and has since become what the community's awesome-hermes-agent directory describes as "the most complete GUI for Hermes." That's a fair description. Chat, terminal, memory browser, skills manager, agent inspector, live streaming of subagent activity—it's all here. Eight themes ship built-in: Official, Classic, Slate, and Mono, each with light and dark variants. The visual design is polished in a way that makes ChatGPT's web interface look dated. Every configuration option Hermes exposes is accessible from within the app—no terminal required once it's set up. The standout feature is the mobile Progressive Web App support via Tailscale, essentially a way of having something that looks and feels like an app on your smartphone. Install the workspace as a web app on your phone, connect through Tailscale, and you have full feature parity with the desktop version from anywhere. You can watch your agents spawn subagents in real time from your couch. That's not a demo feature—it works. The setup is more involved than the others. You need the Hermes gateway running and exposed on port 8642, a configured .env file with API URLs and auth tokens, and optionally a Dashboard API connection for the full sessions, skills, and jobs experience. A Docker Compose file ships with the repo and handles most of this automatically—pull the pre-built images and run docker compose up—but there's still real configuration work involved. As with other GUIs, the instructions are in the repository, but these are the basic steps: For the one-liner install path: curl -fsSL https://raw.githubusercontent.com/outsourc-e/hermes-workspace/main/install.sh | bash That installs Hermes and the workspace together. For manual setup if you already have Hermes running: git clone https://github.com/outsourc-e/hermes-workspace.gitcd hermes-workspace pnpm install cp .env.example .env echo 'HERMES_API_URL=http://127.0.0.1:8642' >> .env pnpm dev Open http://localhost:3000 and complete the onboarding flow. Expect to spend an hour on configuration if you want all the enhanced features unlocked. Tip: Once you have installed and configured the Workspace, you can also ask Hermes to generate executables so you don't have to deal with commands every single time One last thing worth knowing Any of these interfaces will make working with Hermes a truly pleasant experience. Working with an agent doesn’t have to be cumbersome or scary. ChatGPT, Claude, Gemini and all the other chatbots out there are great at coding and produce nice visuals, but their own interfaces are not too polished. Some of these are actually nicer than what those AI behemoths offer. You don't have to figure any of these configurations alone. The Github repositories have all the information you need to set up your agent and almost all the time a simple copy/paste will work. Alternatively, just give Hermes the URL of whichever repo you want to install and ask it to walk you through the setup. Hand it the documentation, describe what you're stuck on, and let it work. It'll cost you some tokens but things usually end up working out. It's almost certainly worth it. Hermes crossed 100,000 GitHub stars in 10 weeks—and the skill system means the agent you use to install one of these GUIs today will be better at it than the one you started with. That's kind of the point.

資料狀態✓ 已擷取全文閱讀原文（Decrypt）

🔍歷史類似事件· 關鍵字 + 標的比對6 則

2026-04-15

你因為擔心監控而改用 Claude，現在它卻要求提供你的護照

相似度 150%關鍵字 claude/you/now

2026-05-01

Grok、ChatGPT、Claude — 11 款 AI 模型預測 Bitcoin 將於 2026 年底達到 $84K 至 $118K

相似度 130%關鍵字 claude/chatgpt

2026-05-01

Binance 研究發現，AI 在利用智慧合約漏洞方面的能力是偵測漏洞的 2 倍

相似度 130%關鍵字 better/than

2026-04-29

機構資金正湧向 Bitcoin，但 Adam Back 表示其速度比你想像的要慢

相似度 100%關鍵字 than/you

2026-04-28

Gemini 推出 Agentic Trading：讓 Claude、ChatGPT 直接在受監管交易所下單，執行策略交易

相似度 100%關鍵字 claude/chatgpt

2026-04-25

Elon Musk 的 XChat 登上 App Store 榜首，發布即超越 ChatGPT 與 Claude

相似度 100%關鍵字 claude/chatgpt

💡 目前用關鍵字 + 標的比對（MVP）· 之後會升級為 embedding 語意搜尋

原始資訊

ID：d3a1f2b2d9

來源：Decrypt

發佈：2026-05-03 18:55:35

分類：一般 · 導出分類 neutral

標的：未指定

社群投票：+0 / −0 · ⭐ 0 重要 · 💬 0 留言