安装
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
这时hermes安装脚本会检查电脑上已经安装的脚本和工具，包括uv(python)、git、nodejs、ripgrep等等。
┌─────────────────────────────────────────────────────────┐
│ ⚕ Hermes Agent Installer │
├─────────────────────────────────────────────────────────┤
│ An open source AI agent by Nous Research. │
└─────────────────────────────────────────────────────────┘
✓ Detected: linux (debian)
→ Checking for uv package manager...
✓ uv found (uv 0.10.8)
→ Checking Python 3.11...
✓ Python found: Python 3.11.2
→ Checking Git...
✓ Git 2.39.2 found
→ Checking Node.js (for browser tools)...
✓ Node.js v22.19.0 found
→ Checking ripgrep (fast file search)...
✓ ripgrep 13.0.0 found
→ Checking ffmpeg (TTS voice messages)...
✓ ffmpeg N-86111-ga441aa90e8-static found
...在安装dependencies的时候可能会花不少时间，具体界面如下：
✓ Virtual environment ready (Python 3.11)
→ Installing dependencies...
→ Some build tools may be needed for Python packages...
✓ Build tools installed
✓ Main package installed
✓ All dependencies installed
→ Installing Node.js dependencies (browser tools)...配置Hermes
在这里选择快速配置。
How would you like to set up Hermes?
↑↓ navigate ENTER/SPACE select ESC cancel
→ (●) Quick setup — provider, model & messaging (recommended)
(○) Full setup — configure everything
由于kimi有多个版本的api url，所以在下面我们选择Custom endpoint
Select provider: [0/2471]
↑↓ navigate ENTER/SPACE select ESC cancel
(○) Nous Portal (Nous Research subscription)
(○) OpenRouter (100+ models, pay-per-use)
(○) Vercel AI Gateway (200+ models, $5 free credit, no markup)
(●) Anthropic (Claude models — API key or Claude Code) ← currently active
(○) OpenAI Codex
(○) Xiaomi MiMo (MiMo-V2 models — pro, omni, flash)
(○) NVIDIA NIM (Nemotron models — build.nvidia.com or local NIM)
(○) Qwen OAuth (reuses local Qwen CLI login)
(○) GitHub Copilot (uses GITHUB_TOKEN or gh auth token)
(○) GitHub Copilot ACP (spawns `copilot --acp --stdio`)
(○) Hugging Face Inference Providers (20+ open models)
(○) Google AI Studio (Gemini models — native Gemini API)
(○) Google Gemini via OAuth + Code Assist (free tier supported; no API key needed)
(○) DeepSeek (DeepSeek-V3, R1, coder — direct API)
(○) xAI (Grok models — direct API)
(○) Z.AI / GLM (Zhipu AI direct API)
(○) Kimi Coding Plan (api.kimi.com) & Moonshot API
(○) Kimi / Moonshot China (Moonshot CN direct API)
(○) MiniMax (global direct API)
(○) MiniMax China (domestic direct API)
(○) Alibaba Cloud / DashScope Coding (Qwen + multi-provider)
(○) Ollama Cloud (cloud-hosted open models — ollama.com)
(○) Arcee AI (Trinity models — direct API)
(○) Kilo Code (Kilo Gateway API)
(○) OpenCode Zen (35+ curated models, pay-as-you-go)
(○) OpenCode Go (open models, $10/month subscription)
(○) AWS Bedrock (Claude, Nova, Llama, DeepSeek — IAM or API key)
→ (○) Custom endpoint (enter URL manually)
(○) Configure auxiliary models...
(○) Leave unchanged然后输入对应的url和key就可以正常使用hermes了。
关于proxy
由于本人电脑长期配置proxy，在运行hermes时反而会出错，具体出错信息如下：
Initializing agent...
Failed to initialize agent: Unknown scheme for proxy URL URL('socks://127.0.0.1:2080/')于是我在bashrc中配置如下：
echo "using hermes"
hermes() {
(unset HTTP_PROXY HTTPS_PROXY ALL_PROXY http_proxy https_proxy all_proxy; exec hermes "$@")
}
在命令行输入type -a hermes
hermes is a function
hermes ()
{
( unset HTTP_PROXY HTTPS_PROXY ALL_PROXY http_proxy https_proxy all_proxy;
exec hermes "$@" )
}
hermes is ~/.hermes/hermes-agent/venv/bin/hermes
这次hermes就不再出现proxy导致的错误了，同时实测使用web search并不会因为unset http_proxy而无法获得google github等网站的内容。
使用感受
hermes在运行速度和使用体验感上让人觉得AI味道很重，确实是一个半成品，在各种edge case程序都可能运行不如意。并且图标等等也是一股AI味道，相比之下claude code就相应收敛的很多，可能hermes在跨平台消息，和记忆上有一点优势，但与此同时我更新了最新的claude，发现其也更新了agent、memory等功能。
同时使用web search等各种其他tools时我发现当AI能够掌握的工具越多，能够访问的网站越多，在当下这个时间节点上，并不代表AI更强大。相反将对应所需要的信息指定清楚，控制上下文长度，能让现在水平有限的AI发挥出更大的能力。