Hermes Agent：自我进化的跨平台 AI 智能体运行时

Hermes Agent, built by Nous Research, is an AI agent runtime that improves itself over time. It creates skills from experience, refines them during use, and persists knowledge across sessions through periodic nudges. It runs on a $5 VPS, a GPU cluster, or serverless infrastructure that costs nearly nothing when idle. You can interact with it via Telegram, Discord, Slack, WhatsApp, Signal, or CLI — all from a single gateway process.

Most AI agents are stateless: they start fresh every session, forget past context, and never improve. Hermes Agent replaces this with a built-in learning loop. It maintains agent-curated memory with periodic nudges; autonomously creates skills after complex tasks; skills self-improve during use; and it uses FTS5 session search with LLM summarization for cross-session recall. The Honcho dialectic user modeling builds a deepening model of who you are across sessions. It is compatible with the agentskills.io open standard.

Hermes Agent works with any model provider: Nous Portal, OpenRouter (200+ models), NovitaAI, NVIDIA NIM, Xiaomi MiMo, z.ai/GLM, Kimi/Moonshot, MiniMax, Hugging Face, OpenAI, or your own endpoint. Switching models requires no code changes — just use the /model command. The Nous Portal subscription bundles tool gateway services (web search, image generation, TTS, cloud browser) under one account, eliminating the need to collect separate API keys for each capability.

Hermes Agent supports six terminal backends: local, Docker, SSH, Singularity, Modal, and Daytona. Modal and Daytona offer serverless persistence — the agent's environment hibernates when idle and wakes on demand, costing nearly nothing between sessions. The same agent runs on a $5 VPS, a Docker container, an SSH server, or a GPU cluster. Native Windows, Linux, macOS, and Android/Termux are all supported with a single-line installer.

Install on Linux/macOS/WSL2/Termux: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash. On Windows (native PowerShell): iex (irm https://hermes-agent.nousresearch.com/install.ps1). The installer handles uv, Python 3.11, Node.js, ripgrep, ffmpeg, and a portable Git Bash. After installation, run source ~/.bashrc then hermes to start chatting. Key commands include hermes model to choose provider, hermes tools to configure tools, hermes gateway to start the messaging gateway, and hermes setup for the full configuration wizard.

Hermes Agent provides two entry points: the terminal UI via hermes, and the messaging gateway via hermes gateway which connects to Telegram, Discord, Slack, WhatsApp, Signal, and Email. Once in a conversation, many slash commands are shared: /model to change the model, /new to start a fresh conversation, /retry to undo the last turn, /skills to browse or invoke skills, and /compress to manage context usage. The gateway allows platform-specific status checks with /status and home setting with /sethome.

Hermes Agent includes a built-in cron scheduler for unattended automations (daily reports, weekly audits) delivered to any platform. It can spawn isolated subagents for parallel workstreams, and write Python scripts that call tools via RPC, collapsing multi-step pipelines into zero-context-cost turns. The skills system serves as procedural memory: skills can be created autonomously, shared from the Skills Hub, or written by the user. The agent supports batch trajectory generation and compression for training future tool-calling models.

Hermes Agent is suitable for developers and teams who need a persistent, learning AI agent that can operate across multiple platforms without a dedicated GPU. It shines in personal productivity (automated reports, cross-session memory), research (batch trajectory generation), and team collaboration (shared agent on messaging platforms). Current caveats: the browser-based dashboard chat pane requires WSL2 on Windows (uses a POSIX PTY). Android/Termux support is documented but tested via a manual path, and the full .[all] extra currently pulls Android-incompatible voice dependencies. The installer is one-line but requires trust in the Nous Research domain and relies on curl/irm.