# Hermes Agent — Self-Improving Multi-Channel AI Agent

> Hermes Agent is a self-improving CLI and gateway agent with memory, skills, cron, and multi-provider support for shared terminal and chat workflows.

## Install

Copy the content below into your project:

## Quick Use

1. Install on Linux/macOS/WSL2:
   ```bash
   curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
   source ~/.zshrc
   ```
2. Start the CLI and run guided setup:
   ```bash
   hermes
   hermes setup
   ```
3. Verify:
   - Run `hermes model` and `hermes tools`, then finish one conversation from the terminal.

## Intro

Hermes Agent is a self-improving CLI and gateway agent with memory, skills, cron, and multi-provider support for shared terminal and chat workflows.

- **Best for:** operators who want one open agent to span CLI, cron, and messaging channels
- **Works with:** Python 3.11, Node.js, ripgrep, ffmpeg, Telegram/Slack/Discord gateways, MCP servers
- **Setup time:** 15-30 minutes

## Practical Notes

- Quant: the installer wires Python 3.11, Node.js, ripgrep, ffmpeg, and a portable Git Bash on native Windows.
- Quant: the docs call out 40+ tools and 7 terminal backends, useful when you need a remote-first agent footprint.

## Why it matters

Hermes is interesting because it treats agent memory and channel delivery as first-class runtime features instead of add-on scripts.

- The same agent can live in terminal, Telegram, Discord, Slack, WhatsApp, Signal, and email without forcing a separate automation stack.
- It exposes operational commands such as `hermes model`, `hermes tools`, `hermes gateway`, and `hermes doctor`, which lowers day-2 maintenance cost.
- The README explicitly positions skills, memory search, and scheduled automations as part of the core loop, not optional demos.

## Rollout pattern

- Start with the CLI only and pin one provider + one toolset before you expose any gateway channel.
- Add cron or gateway delivery only after you can replay the same task three times from the terminal with consistent evidence.
- Use separate environments for experiments versus production because the agent can execute tools and accumulate long-lived memory.

## Watchouts

The project is powerful enough to touch local systems and remote chats, so approval gates and environment separation matter more here than in a prompt-only assistant.

### FAQ

**Q: Does it lock me to one model vendor?**
A: No. The README lists OpenAI, OpenRouter, NVIDIA NIM, GLM, Kimi, MiniMax, Hugging Face, and custom endpoints.

**Q: Can I start from chat instead of the terminal?**
A: Yes, but the terminal path is the cleanest first verification route before enabling gateways.

**Q: What is the fastest smoke test?**
A: Install, run `hermes setup`, pick a model, and complete one tool-free CLI conversation before enabling extra tools.

## Source & Thanks

> Source: https://github.com/NousResearch/hermes-agent
> License: MIT
> GitHub stars: 146,285 · forks: 22,902

---

<!-- ZH -->

## 快速使用

1. 在 Linux/macOS/WSL2 安装：
   ```bash
   curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
   source ~/.zshrc
   ```
2. 启动 CLI 并完成向导：
   ```bash
   hermes
   hermes setup
   ```
3. 验证：
   - 运行 `hermes model` 与 `hermes tools`，并在终端完成一次完整对话。

## 简介

Hermes Agent 把 CLI、消息网关、长期记忆、技能系统与定时任务放进同一套开源代理里，并支持多模型切换，适合希望把终端工作流与聊天入口统一到一个可持续运行的 AI Agent 的团队。

- **适合谁：** 希望用一套开源 Agent 同时覆盖 CLI、定时任务与消息通道的团队
- **可搭配：** Python 3.11、Node.js、ripgrep、ffmpeg、Telegram/Slack/Discord 网关与 MCP servers
- **准备时间：** 15-30 分钟

## 实战建议

- 量化信息：安装脚本会处理 Python 3.11、Node.js、ripgrep、ffmpeg，Windows 还会带上便携 Git Bash。
- 量化信息：文档列出了 40+ 工具与 7 种终端后端，适合要把 Agent 跑在远端环境的团队。

## 为什么值得收录

Hermes 的价值在于它把记忆、技能、通道接入和定时执行都做成了运行时原生能力，而不是让你再拼一层胶水脚本。

- 同一个 Agent 能在终端、Telegram、Discord、Slack、WhatsApp、Signal 与邮件之间复用，减少多入口维护成本。
- 像 `hermes model`、`hermes tools`、`hermes gateway`、`hermes doctor` 这样的运维命令已经内建，便于长期管理。
- README 明确把技能、历史检索和定时自动化当成核心能力，而不是只给一个演示型聊天壳子。

## 落地路径

- 先只开 CLI，固定 1 个模型 provider 与 1 套工具集，再考虑开放消息通道。
- 当你能在终端把同一任务稳定复现 3 次之后，再接入 cron 或外部网关。
- 测试环境与生产环境分开，因为它既能执行工具，也会累积长期记忆。

## 注意事项

这类“可执行 + 可跨通道”的 Agent 比普通提示词系统风险更高，权限闸门、日志证据和环境隔离都应该更严格。

### FAQ

**会绑定某一家模型厂商吗？**
答：不会。README 列出了 OpenAI、OpenRouter、NVIDIA NIM、GLM、Kimi、MiniMax、Hugging Face 和自定义端点。

**能直接从聊天端开始吗？**
答：可以，但第一次验证更建议从 CLI 开始，问题更容易定位。

**最快的冒烟验证是什么？**
答：安装后跑 `hermes setup`，选好模型，再完成一次不带外部工具的 CLI 对话。

## 来源与感谢

> Source: https://github.com/NousResearch/hermes-agent
> License: MIT
> GitHub stars: 146,285 · forks: 22,902


---
Source: https://tokrepo.com/en/workflows/hermes-agent-self-improving-multi-channel-ai-agent
Author: Agent Toolkit