Hermes Agent explained: skills, plugins and how to set it up
Short answer
As of October 2026, Hermes Agent is a free, open-source (MIT) personal AI agent from Nous Research, launched February 25, 2026, with version 0.21.5 released September 24, 2026. You extend it with skills (SKILL.md instruction files, more than 50 bundled and 150+ official optional ones), Python plugins and MCP servers. The agent costs nothing; you pay for the AI model you connect, or use a Nous Portal plan from $0 to $200 a month.
Based on the official documentation as of October 2026, this guide explains Hermes Agent. It covers what its add-ons are called and how to set it up. The Hermes Agent skills directory lists the add-ons one by one.
What Hermes Agent is
Hermes Agent is free, open-source agent software from Nous Research. It runs on your computer or a server and works from a terminal, a desktop app or chat apps like Telegram and Slack. It is not an AI model. You connect any model, and Hermes gives it tools, memory and skills.
Nous Research announced it on February 25, 2026. It called it "an autonomous agent that lives on your server, remembers what it learns, and gets more capable the longer it runs." The GitHub repo is MIT licensed and showed about 239,000 stars in early October 2026.
Its pitch is a "built-in learning loop": after a hard task, it can save what worked as a reusable skill and improve it later.
The newest stable release on October 1, 2026 was v0.21.5, published September 24, 2026. It added a Desktop plugin SDK, replaced the MCP tab with a Connectors page, and added official Blender Lab and NVIDIA app/Broadcast plugins.
The parts, defined plainly
Hermes has three kinds of add-ons: skills, plugins and MCP servers. Skills are instruction files the agent reads when a task needs them. Plugins are Python code that adds tools, memory or model providers. MCP servers are outside tool services. Built-in tools and memory sit underneath all three.
| Term | What it means |
|---|---|
| Skill | A folder with a SKILL.md file of step-by-step instructions, sometimes with scripts. The agent loads it only when relevant. |
| Bundled skill | One of more than 50 skills that ship with every install, copied into ~/.hermes/skills/. |
| Official optional skill | A Nous-maintained skill that is off until you install it. There are more than 150. |
| Plugin | Python code that adds tools, hooks, slash commands, memory backends, context engines or model providers. |
| Memory provider | A plugin that adds long-term memory, such as Honcho, Mem0, Supermemory or the local Holographic store. Only one external provider can be active. Built-in memory (MEMORY.md and USER.md) always stays on. |
| MCP | Model Context Protocol, an open standard for connecting agents to outside tool servers. |
| Gateway | The part that connects Hermes to chat apps like Telegram, Discord and WhatsApp. |
| Nous Portal | Nous Research's subscription that bundles many models and managed tools under one login. |
Skills
The skills docs call them "on-demand knowledge documents" and say they follow the agentskills.io open standard. Anthropic originally developed that format, and Claude, ChatGPT/Codex and OpenClaw support it too.
Hermes reads only each skill's name and description until a task needs the full file, which saves tokens. The agent can also write and edit its own skills with a tool called skill_manage. You can also type /learn and point it at reference material, and Hermes writes a skill from it.
A built-in Curator tidies up skills the agent wrote itself. It is on by default. Skills unused for 30 days move to ~/.hermes/skills/.archive/, and hermes curator restore <skill> brings one back. Run hermes curator pin <skill> to keep one, or set curator.enabled: false to turn it off.
The Skills Hub installs from official optional skills, GitHub repos (including OpenAI, Anthropic, Hugging Face and NVIDIA), skills.sh, ClawHub, LobeHub and direct URLs. To add one by hand, drop a SKILL.md into ~/.hermes/skills/.
Plugins
The plugins docs list four types: general plugins, memory providers, context engines and model providers. General plugins are disabled by default. They load only after you enable them. One bundled example is disk-cleanup, which clears temporary files.
Find plugins in the Plugin Catalog. It is a human-reviewed list labeled Official (maintained by Nous Research) or Community. Each entry is pinned to an exact code version and security-scanned. Install one by name with hermes plugins install <name>. Hermes Desktop also accepts one-click hermes:// install links.
MCP servers and built-in tools
Hermes can connect to MCP servers over a local process or HTTP, with OAuth sign-in. It can also act as an MCP server for other agents with hermes mcp serve. Built-in tools cover web search, a terminal, files, a browser, image generation, scheduled tasks and memory.
How to install it and what you need
You install Hermes with one command or a desktop app. No paid plan is required. You do need a model. That can be an API key, a Nous Portal account, or a local model with at least a 64,000-token context window.
The installation guide gives these commands:
| Platform | How to install |
|---|---|
| Linux, macOS, WSL2 | curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash |
| Windows (PowerShell) | iex (irm https://hermes-agent.nousresearch.com/install.ps1) |
| Desktop app | Download for macOS (Apple Silicon) or Windows from the Hermes site |
| Android | The Termux package on aarch64 devices |
Intel Macs are not supported. First-party installs run on Python 3.14 and need Git, curl and tar.
After installing, reload your shell. Run hermes setup --portal for quick setup, or hermes model to pick another provider. hermes gateway setup connects chat apps, and hermes doctor helps if something breaks.
Skills install with hermes skills install official/<category>/<skill>. Plugins install with hermes plugins install <name> (or owner/repo for one outside the catalog), then hermes plugins enable <name>. Memory backends are chosen with hermes memory setup. New to local models? Start with our Ollama guide.
What it costs
Hermes Agent itself is free. The FAQ says "You pay only for the LLM API usage from your chosen provider." Your real bill depends on the model, how much you use it, and whether you rent a server or managed hosting.
| Option | Price as of October 2026 |
|---|---|
| Hermes Agent software | Free (MIT license) |
| Local model (Ollama, llama.cpp, vLLM) | $0 beyond your own hardware |
| Nous Portal Free | $0/month, free models only |
| Nous Portal Plus | $20/month, includes $22 of credits |
| Nous Portal Super | $100/month, includes $110 of credits |
| Nous Portal Ultra | $200/month, includes $220 of credits |
| Hermes Cloud (managed hosting) | From $0.56/day running, $0.03/day stopped |
On Hermes Cloud, model and tool use bill on top of the instance price. Nous Portal is what Nous calls "the recommended way to run Hermes Agent." It bundles many models plus managed tools such as web search.
If you host it yourself, Hostinger (which sells hosting) estimates $4 to $25 a month for a server. Model fees add about $2 to $60 a month. For a wider view, see local AI vs subscription cost and AI chatbot prices compared.
Who it is for, and its limits
Hermes suits people who want an always-on assistant they control, with any model and their own data. It runs real commands on your machine, so some comfort with a terminal helps. Casual chat users will find ChatGPT or Claude easier, since those need no setup or server.
The messaging docs list more than 25 chat platforms, including Telegram, Discord, Slack, WhatsApp, Signal and Microsoft Teams.
Limits to know:
- Setup is technical. You manage keys, config files and updates yourself unless you pay for Hermes Cloud.
- Quality depends on the model you pick. Small local models may struggle with long, multi-step tasks.
- Hermes collects no telemetry, but every prompt goes to whichever model provider you configure.
Safety: what to check before adding skills and plugins
Hermes can run shell commands, edit files, browse the web and send messages for you. Plugins run as ordinary Python with full access to your account. Install only what you trust, keep command approvals on, run the terminal in a sandbox, and lock down who can message your bot.
Practical steps from the security, plugins and skills docs:
- Keep approvals on. The default "smart" mode uses a second model to judge risky commands. Switch to "manual" to approve every one. Avoid
--yolooutside a sandbox. - Treat plugins as trusted code. The docs warn: "Plugins run as regular in-process Python: a malicious plugin can ignore every gate here." Pin installs to a commit with
--ref. When an install asks you to grant a capability, the docs call that "a statement of trust in the plugin author", not a code audit. - Read community skills first. Hub skills are scanned for data theft, prompt injection and destructive commands. A "dangerous" verdict blocks install, even with
--force. A scan is not a guarantee. - Review skills the agent writes. Set
skills.write_approval: trueto approve each skill the agent creates before it is saved. - Use a sandbox. Run the terminal in Docker (
terminal.backend: docker) and as a non-root user. - Lock the chat gateway. It denies unknown users by default. Use DM pairing or allowlists, and never set
GATEWAY_ALLOW_ALL_USERS=true. - Watch money and dual-use skills. Payment skills in the optional catalog, such as stripe-link-cli and mpp-agent, can spend real money. Skills such as godmode, scrapling and sherlock should be used only where legal and authorized.
Coming from OpenClaw
Hermes can import an OpenClaw setup with one command, hermes claw migrate. The migration guide says it brings over your persona, memory, skills, settings and chat app tokens. Preview first with a dry run, because some items, such as WhatsApp pairing, must be redone by hand.
Run hermes claw migrate --dry-run, then hermes claw migrate. Add --preset full --migrate-secrets to bring API keys. Imported skills land in ~/.hermes/skills/openclaw-imports/. Read what OpenClaw is for background.
Related guides
Hermes is one of several agents that grow through add-ons. These guides explain how ChatGPT, Claude and OpenClaw handle skills, plugins and connectors. The Hermes directory lists official skills and plugins with a one-line summary each. The Mac guide helps if you plan to run models locally.
- ChatGPT vs Claude vs Hermes vs OpenClaw vs Muse compared
- Hermes Agent skills and plugins directory
- Claude skills, plugins and connectors explained
- ChatGPT apps, GPTs and skills explained
- What is OpenClaw?
- Which Mac for local AI
Sources
- Nous Research releases (October 2026)
- NousResearch/hermes-agent on GitHub (October 2026)
- Hermes Agent v0.21.5 release notes (October 2026)
- Hermes docs: Installation (October 2026)
- Hermes docs: Skills System (October 2026)
- Hermes docs: Plugins (October 2026)
- Hermes docs: Plugin Catalog (October 2026)
- Hermes docs: Curator (October 2026)
- Hermes docs: Memory Providers (October 2026)
- Hermes docs: Bundled Skills Catalog (October 2026)
- Hermes docs: Optional Skills Catalog (October 2026)
- Hermes docs: MCP Integration (October 2026)
- Hermes docs: Tools and Toolsets (October 2026)
- Hermes docs: Messaging Platforms (October 2026)
- Hermes docs: Security (October 2026)
- Hermes docs: FAQ (October 2026)
- Hermes docs: Nous Portal (October 2026)
- Hermes docs: Migrate from OpenClaw (October 2026)
- Hermes Cloud and Nous Portal plans (October 2026)
- Agent Skills overview (agentskills.io) (October 2026)
- Hostinger: Hermes Agent cost (October 2026)
Frequently asked questions
Is Hermes Agent the same as the Hermes AI models?
No. Hermes 4 and Hermes 4.3 are language models Nous Research released in 2025. Hermes Agent is separate software that wraps any model, including Claude, GPT, Gemini, a Hermes model or a local model, and gives it tools, memory and skills.
Is Hermes Agent free?
Yes. The FAQ says Hermes Agent is free and open-source under the MIT license, and you pay only for model usage from your chosen provider. A local model through Ollama or llama.cpp costs nothing beyond your own hardware.
Can Hermes Agent use Claude or ChatGPT skills?
Often, yes. Hermes skills follow the open Agent Skills (agentskills.io) format that Anthropic created and that Claude, ChatGPT/Codex and OpenClaw also support. Hermes trusts repos such as openai/skills and anthropics/skills by default and can install from them.
Does Hermes Agent run on an Intel Mac?
No. The installation docs say the macOS installer is Apple Silicon only and Intel Macs are not a supported platform. Linux, WSL2, native Windows and Android (Termux) are supported.
Can I move from OpenClaw to Hermes Agent?
Yes. The command hermes claw migrate imports your OpenClaw persona, memory, skills, settings and messaging tokens. Run it with --dry-run first to preview what will change.