使用 Homebrew, winget, apk, dnf 安装 hermes-agent

项目链接

• 主页
• 仓库

命令和别名

• hermes
• hermes-acp
• hermes-agent

项目历史

The GitHub repository was created on July 22, 2025. The README describes Hermes as a self-improving AI agent with a built-in learning loop that creates skills from experience, improves them during use, searches past conversations, and builds cross-session user context.

By 2026, the documentation and README described a broad runtime rather than a narrow terminal chatbot: a full TUI, model-provider switching, messaging gateways, scheduled automations, subagents, terminal backends, memory, MCP integrations, and research-oriented trajectory generation.

采用历史

Hermes Agent sits in the post-2024 generation of coding and task agents where the package is expected to install not only an executable, but also a managed runtime, provider configuration, skills, memory, and optional gateways. Its installer documentation says it handles Python, Node.js, ripgrep, ffmpeg, repository checkout, virtual environment, and global command setup.

The project's packaging story includes multiple command names in Homebrew and a managed install layout under `~/.hermes`, which reflects the way agent CLIs blurred the line between a local package and a persistent personal runtime.

使用方式

A user typically starts with `hermes`, chooses or configures a model provider, and then works through a terminal UI or messaging gateway. The README documents provider flexibility through Nous Portal, OpenRouter, OpenAI, user endpoints, and other backends.

Skills are central to day-to-day use. Official docs describe user and agent-created skills under `~/.hermes/skills/`, a bundled skills catalog, hub-installed skills, and a write-approval gate for users who want review before background self-improvement writes skill changes.

为什么软件包爱好者会关心

Hermes Agent is package-nerd interesting because it makes a local CLI behave like an extensible operating environment: it owns configuration, skills, memory, provider keys, optional messaging integrations, and tool execution backends. Installing the package is therefore closer to bootstrapping a development workstation than installing a single-purpose Unix command.

It also shows how AI-agent distributions began to depend on language-package managers and runtime bootstrapping. The installer leans on Python and Node ecosystems while exposing a single `hermes` command, making the package a wrapper around a larger managed environment.

时间线

• 2025-07-22: Public GitHub repository creation date recorded by GitHub.
• 2026: Documentation described installer-managed dependencies, managed checkout, virtual environment, and global `hermes` command setup.
• 2026: Official skills documentation described background self-improvement review and optional write approval for skill changes.

Related projects

• Hermes Agent is related to autonomous coding and task agents such as Claude Code, Codex-style terminal agents, OpenClaw, and other tool-calling shells. The README explicitly documents migration from OpenClaw.
• Its skills model is related to agentskills.io and to the broader agent-memory pattern in which reusable procedures are stored outside the prompt and loaded only when relevant.

来源

• GitHub README for scope, install model, provider support, and migration notes.
• Official documentation home and installation page for product positioning and installer behavior.
• Official skills and skills-catalog pages for self-improvement and bundled skill behavior.