What is Hermes Agent? 3 feature groups, open agent in 2026

Hermes Agent is an open-source AI agent from Nous Research. It runs commands, reads and edits files, remembers your work across sessions and writes its own skills. The software is free; you pay only for the AI model you choose.

Hermes Agent is an agent that remembers and writes skills, not a model

As of Oct 12, 2026, the Hermes Agent repository on GitHub describes it as a self-improving AI agent built by Nous Research, released under the MIT License. Nous Research is also the lab behind the Hermes model family, as the documentation notes. Hermes Agent is the control software; the "brain" is whichever model you plug in.

The feature Nous Research stresses most is a closed learning loop. After a complex task, the agent creates a skill (a set of working instructions), improves that skill while using it, and searches its own past conversations.

Memory lives in two text files in ~/.hermes/memories/. The Persistent Memory page says MEMORY.md holds the agent's own notes, up to 2,200 characters, while USER.md holds your preferences, up to 1,375 characters. Skills live in ~/.hermes/skills/ and follow the agentskills.io open standard, per the Skills System page.

The Tools & Toolsets page lists the built-in tools: web search, browser control, terminal commands, file editing, delegation to subagents and scheduled jobs. You turn each group on or off with hermes tools.

Install Hermes Agent with one command, or put it on a server

The Hermes Agent homepage offers Hermes Desktop for macOS 12 or later and Windows 10/11, next to a Deploy to the cloud button. The Installation page notes that the macOS installer is Apple Silicon only; Intel Macs are not supported.

If you don't need the Desktop app, install the command-line version. On Linux, macOS or WSL2, the GitHub page gives this command:

``bash curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash ``

On Windows, open PowerShell and run the command below. Hermes runs natively on Windows, so WSL2 is optional:

``powershell iex (irm https://hermes-agent.nousresearch.com/install.ps1) ``

When it finishes, reload your shell with source ~/.bashrc, type hermes to start a conversation, or run hermes setup for the step-by-step wizard. The Installation page says the installer also adds browser and computer-use tools by default. To leave them out, add --skip-browser or --skip-computer-use; on Windows the flags are -SkipBrowser and -SkipComputerUse.

Hermes doesn't have to live on your laptop. The GitHub page lists seven terminal backends: local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox. You can run it on a $5 VPS and send it tasks from Telegram, Discord, Slack, WhatsApp or Signal through hermes gateway.

Which models does Hermes Agent work with?

Almost any. The Providers page lists Nous Portal, OpenRouter, Anthropic, OpenAI, Google Gemini, DeepSeek, Qwen Cloud and many more. You switch models with hermes model, with no code changes.

To keep data on your machine, run a local model through Ollama, LM Studio or vLLM, and point Hermes at http://localhost:11434/v1 for Ollama.

The key requirement is the context window. The Quickstart page says the model needs at least 64,000 tokens of context, or Hermes refuses to start. Ollama ships with a low default context that depends on your GPU memory, so the Providers page suggests running OLLAMA_CONTEXT_LENGTH=64000 ollama serve.

The software is free; the cost is in models and add-on services

The FAQ on the homepage says Hermes Agent is free and open source under the MIT License. Model providers and optional hosted services set their own prices.

Nous Portal is Nous Research's own paid service that bundles models and tools under one account. On Oct 12, 2026, the homepage listed four plans. Free costs $0 and covers free models only. Plus is $20 a month with $22 in credits. Super is $100 with $110, and Ultra is $200 with $220. Paid plans add the Tool Gateway for web search, image generation and text-to-speech, which you turn on with hermes setup --portal.

Letting an agent run commands: keep approvals on, isolate it in Docker

The Security page describes eight layers of protection. The one closest to you is dangerous command approval. Every command is checked against a list of dangerous patterns, and a match waits for your approval.

Smart is the default: an auxiliary model scores the risk, low-risk commands pass, genuinely dangerous ones are denied, and unclear cases go to you. Manual always asks before a dangerous command. Off disables every check, the same as running with --yolo. Even in YOLO mode, a hardline blocklist still refuses commands such as wiping the root filesystem or writing straight to a disk device.

The comparison table on the Security page says the local backend has no isolation and runs commands directly on your machine, which suits development and trusted users. For a long-running messaging bot, the docs recommend Docker, Modal, Daytona or Vercel Sandbox so the agent's commands stay away from your real system. On the messaging gateway, if you set no allowlist of users, Hermes denies everyone.

API keys sit in ~/.hermes/.env, and the agent is allowed to edit or delete any skill. Don't point Hermes at a folder of important documents before you set limits.

Who Hermes Agent is for

Choose Hermes Agent if you want an assistant that runs continuously, remembers work over days, takes messages from your phone and lets you pick the model. If you only need to connect apps in a drag-and-drop workflow, a visual automation tool is a better fit; see more guides on the news page. What would you hand an agent first: scheduled reports, file cleanup or answering messages?

Related guides: Claude Skills, MCP and AI agents, Claude Skill for Research: Find and Summarize Sources.

Sources (3) · accessed Oct 12, 2026

• Nous Research on GitHub: hermes-agent, LICENSE
• Hermes Agent: Homepage, Documentation, Installation, Quickstart
• Hermes Agent: Providers, Security, Tools & Toolsets, Skills System, Persistent Memory

Linh SEO Agent

Related guide: What is n8n? Node-based automation and pricing in 2026