What Is Hermes Agent? How to Install It and What It Costs
Install commands, the Windows and Mac requirements for the desktop app, Nous Portal prices and the 64,000-token context minimum for local models.
In short: Hermes Agent is an open-source, MIT-licensed AI agent built by Nous Research. It keeps memory across sessions, writes its own skill files after finishing complex tasks, and runs from a terminal, a desktop app or messaging apps such as Telegram and Slack. The software itself is free: you pay the model provider you choose, or for an optional Nous Portal plan.
Hermes Agent is not a language model. It is agent software that connects the model you pick to tools, memory and messaging channels. What sets it apart from an ordinary chat app is that it saves how it did a finished task as a skill and remembers it in the next session. The software is free, so what it costs depends on the model you choose. You can install it with a desktop app or a one-line command, but the first setup asks for two decisions: whether the model runs in the cloud or on your own machine, and which data the agent can reach.
What is Hermes Agent?
#Hermes Agent is an AI agent that Nous Research publishes on GitHub under the MIT license, and it improves itself as you use it. The project's README on GitHub introduces it in its first line as the self-improving AI agent built by Nous Research. Self-improving means two things here: the agent keeps memory across sessions, and after it finishes a complex task it writes down how the task was done in a skill file, which it updates in later runs.
There is no language model inside Hermes Agent. According to the README, you connect Nous Portal, OpenRouter, OpenAI or your own endpoint, and you can switch models with the hermes model command without changing any code. The agent's job is to carry that model's decisions through to tools, memory and channels. We cover the agent concept itself, the loop in which a language model picks the next step and tool on its own, in our post what is an AI agent.
The license is clear: the repository's LICENSE file opens with "MIT License" and "Copyright (c) 2025 Nous Research". The name needs more care, because "Hermes" is used by other software too. Before you download the install script, check that you are on hermes-agent.nousresearch.com or github.com/NousResearch/hermes-agent.
What can Hermes Agent do?
#Hermes Agent works as an agent that runs commands in a terminal for you, answers requests from messaging apps and runs scheduled jobs unattended. The features on the Hermes Agent homepage and in the README group around these three uses.
The most basic use is the terminal. The classic interface opens with hermes and the newer one with hermes --tui, and the agent can run commands for you and show the result. You choose where those commands run: on your local machine, in a Docker container or on a remote server over SSH. The agent can also gather information from the web with web search and browser automation.
The messaging gateway is what takes the agent beyond the terminal. A single gateway process connects to platforms such as Telegram, Discord, Slack, WhatsApp, Signal and email, so you can message the agent from your phone. The same process runs the built-in cron scheduler: a report, backup or daily briefing that you describe in plain language gets done at the time you set, without you being there.
When the work grows, the agent can split it up. For parallel work it can start isolated subagents, each with its own conversation and terminal. Where the built-in tools are not enough, you connect the agent to outside systems by adding an MCP server to ~/.hermes/config.yaml. We explain how the protocol works in what is MCP.
You do not need to turn on all of these layers on day one. The official Quickstart recommends getting one plain chat working first and adding layers such as the gateway, cron and skills later. That is the right order for a company pilot too, because every layer you add widens what the agent can reach.
How do you install Hermes Agent?
#You install Hermes Agent in one of two ways: with the Hermes Desktop app on macOS and Windows, or with a one-line install script on Linux, macOS, WSL2 and Windows. As of November 12, 2026, the latest release is v0.23.1, released on November 10. The Quickstart recommends the desktop app. Its installer sets up the command-line tool and the desktop app together.
How do you install Hermes Desktop on Windows and Mac?
#On a Mac, Hermes Desktop installs from a DMG package. On Windows, you can get it from the Microsoft Store, with the Hermes-Setup.exe installer or with the .appinstaller file. According to the official Installation page, the macOS package runs only on Apple Silicon Macs, and there is no ready-made desktop package for Intel Macs. The docs' Platform Support page shows a way around this: on an Intel Mac you can install the command-line version and open the desktop app with the hermes desktop command.
On Windows, the version matters. According to the Platform Support page, the desktop package (MSIX) needs Windows 11 22H2 or later. If you are on Windows 10, you can install Hermes Agent as a command-line tool with the PowerShell command below, which works on both Windows 10 and 11.
How do you install Hermes Agent from the command line?
#On Linux, macOS and WSL2, Hermes Agent installs with a single install script. The Quickstart sequence looks like this:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
source ~/.bashrc # on zsh: source ~/.zshrc
hermes setup # if you already know your provider: hermes model
hermes --tui # for the classic interface: hermes
hermes --continue # check that your last session comes back
The last command confirms that your previous session was saved and can be restored. According to the Installation page, the script needs Git, curl, tar and SHA-256 utilities on the system. The docs' FAQ & Troubleshooting page says not to start the installer with sudo. If dependencies need elevated rights, an interactive install asks for your password once by itself.
On Windows 10 and 11, Hermes Agent installs without WSL2. This command in PowerShell is enough:
iex (irm https://hermes-agent.nousresearch.com/install.ps1)
If Git is missing, the Windows installer can set up Git for Windows itself. On Android, Hermes Agent installs on aarch64 devices from a separate APT repository built for Termux.
How do you set up Hermes Agent after installing it?
#The hermes setup wizard handles the first setup and offers three modes. Quick Setup signs in to your Nous Portal account with OAuth and bills usage to your Portal subscription. In Full Setup you configure each provider and tool with your own API keys. Blank Slate starts with only the provider, model, file and terminal tools. Web, memory, cron, skills and MCP start switched off in this mode. That makes it the best starting point for a controlled company pilot, because you widen the agent's access one step at a time. If the wizard stops halfway or a setting breaks, the first command in the Quickstart's recovery steps is hermes doctor.
For safety, the Quickstart recommends running the agent's commands in Docker or on a remote server instead of your local machine. Switching to Docker takes one setting:
hermes config set terminal.backend docker
Read the install script before you run it, and do your first test on an account with no access to company data. Because the agent runs commands on your behalf, a mistake or a malicious instruction can touch everything it can reach.
Is Hermes Agent free?
#Yes, the Hermes Agent software itself is free. The FAQ & Troubleshooting page says Hermes Agent is free and open source (MIT), that you pay only for LLM API usage from the provider you choose, and that local models are free to run. There is no separate fee for the desktop app either: the Hermes Desktop page answers yes to the question of whether it is free, both to download and to use.
So what it costs depends on where you run the model.
| Cost item | Who charges? | Note |
|---|---|---|
| Hermes Agent software | Free | MIT license |
| Hermes Desktop | Free | Free to download and use |
| Model usage | The provider you choose | Price on the provider's own page |
| Nous Portal | Nous Research, optional | Plans in the table below |
| Local model | No license fee | Hardware and electricity are your cost |
You can also connect to some providers through a subscription you already have. The Quickstart's example is the OpenAI Codex option, which works with your ChatGPT or Codex subscription.
How much do Nous Portal plans cost?
#Nous Portal is Nous Research's service for model access and hosted tools. As of November 12, 2026, the Hermes Agent homepage lists one free and three paid plans.
| Plan | Price | Models | Cloud hosting (one agent) |
|---|---|---|---|
| Free | $0, pay as you go | Free models only | None |
| Plus | $20 per month | Free and paid models | 4 GB RAM, 25 GB disk |
| Super | $100 per month | Free and paid models | 4 GB RAM, 100 GB disk |
| Ultra | $200 per month | Free and paid models | 8 GB RAM, 100 GB disk |
Paid plans include model usage in Hermes Agent, hosted tools such as web search, image generation, speech and a cloud browser, and one agent running in the cloud. The main difference between the plans is how much usage you get: Super offers 5 times the usage of Plus and Ultra 10 times, but the page gives no exact quota for Plus. Every plan can add pay-as-you-go top-ups. Prices are in US dollars, and the page shows no tax or local-currency prices.
What are Hermes Agent Skills, Desktop and the gateway?
#Skills, Desktop and the gateway are three parts of Hermes Agent built on the same core: a skill holds how the agent does a task, Desktop opens the agent on your desktop, and the gateway takes it to messaging apps.
What are Hermes Agent Skills?
#A skill is a SKILL.md file that walks Hermes through a specific task step by step. According to the Skills System docs, the agent first reads only the short descriptions of its skills and loads the full content when a task needs it. That way, skills it is not using take up no room in the context window. Bundled skills ship in ~/.hermes/skills/. You list more with hermes skills browse and install them from the Skills Hub with hermes skills install, and a security scan runs before each install. The format is compatible with the agentskills.io open standard.
If you plan to write your own skills, design for this step-by-step loading from the start. RuneLab's Agent Skills and Capability Design: Modular Agent Architecture course works on exactly that design: reusable skill packages, progressive context disclosure, scripts bundled with a skill, and loading capabilities when needed so an agent scales within its context window.
What separates a skill from memory is the kind of knowledge it holds. According to the FAQ & Troubleshooting page, memory stores facts and a skill stores how a task is done, and both persist across sessions. Memory remembers which report you want and in what format, and a skill remembers the steps for preparing it.
Because skills change how the agent behaves, treat them like code. According to the Skills System docs, the agent can modify or delete any skill, and a skill you install from outside is a new instruction given to the agent. The security scan is a first filter, so read the content yourself before installing. We explain how instructions in outside content can steer an agent in what is prompt injection. To assess the risk of third-party skills, tools and servers across an organization, RuneLab's Agentic AI Supply Chain and Model Context Protocol (MCP) Security course covers tool poisoning, rug pulls, shadowing and supply chain risks in MCP servers, third-party agent tools and A2A integrations.
What is Hermes Desktop?
#Hermes Desktop is the desktop app for Hermes Agent, and it uses the same agent core as the command line and the gateway. According to the Hermes Desktop docs, settings, API keys, sessions, skills and memory are shared across all three interfaces. So you can continue a task you started in the terminal on the desktop, with the same memory. If you installed the command-line version, you can open the app with the hermes desktop command.
How does the Hermes Agent gateway work?
#The messaging gateway is a single background process that connects to all the platforms you configure, manages sessions and runs cron jobs. According to the Messaging Gateway docs, you set it up with the hermes gateway setup command. By default, the gateway denies users who are not on the allowlist or have not been paired by direct message. The docs call this the safe default for a bot with terminal access and advise against opening the bot to everyone.
Does Hermes Agent work with a local model?
#Yes, Hermes Agent works with a local model or with any OpenAI-compatible server. According to the FAQ & Troubleshooting page, you choose Custom endpoint in hermes model and enter the server's address. This works with Ollama, vLLM, llama.cpp server, SGLang, LocalAI and other OpenAI-compatible servers. The docs' Ollama example looks like this:
hermes model
# Select: Custom endpoint (enter URL manually)
# API base URL: http://localhost:11434/v1
# API key: ollama
# Model name: qwen3.5:27b
# Context length: 64000
How large a context window does a local model need?
#Hermes Agent needs a model with a context window of at least 64,000 tokens. According to the Quickstart, models with smaller windows are rejected at startup. The Context length line in the example tells Hermes the size of the model's window. The server has to run with that window too, and local servers may start with a smaller one by default: according to the Ollama FAQ, Ollama uses 4,096 tokens by default.
In llama.cpp you raise the window with the --ctx-size 65536 option. In Ollama you set the num_ctx parameter, or the OLLAMA_CONTEXT_LENGTH environment variable when you start the server, to 65536. We walk through installing Ollama and running a model on your own computer in our guide how to run an LLM locally. To learn it together with your team, see RuneLab's Running Local LLMs with Ollama course.
Can Hermes Agent connect to an in-house endpoint?
#Yes, the Custom endpoint option works not only for Ollama on a laptop but also for a company's own OpenAI-compatible endpoint. That way every Hermes install on the team points to the same address, and model choice, budgets and logging are managed in one place. RuneLab's RuneGate is that kind of endpoint: it puts your models behind a single OpenAI-compatible endpoint, routes each request to the most cost-effective model that can do the job, stops requests before a team budget is exceeded and logs every request. If data must not leave the company, RuneGate works with local models on RuneBox. RuneBox is on-premise LLM infrastructure installed on the company network. Because data is processed within the organization's boundaries, it makes your BDDK and KVKK assessments easier. Compliance remains your organization's own assessment.
Where does Hermes Agent store data?
#Hermes Agent stores conversations, memory and skills locally, in the ~/.hermes/ folder. According to the FAQ & Troubleshooting page, API calls go only to the provider you configure, and Hermes Agent collects no telemetry or analytics. That does not mean the data always stays on your machine. If you use a cloud model, Nous Portal tools or a messaging platform, the relevant part of the content goes to those services as well.
Running locally does not by itself make a setup compliant with KVKK, Turkey's personal data protection law, either. Check whether the files and messages the agent reads contain personal data and, if so, for what purpose it is processed. If the model provider or messaging platform is abroad, work out how that is assessed as a cross-border data transfer. Also pin down who can see the conversation and memory records under ~/.hermes/, how long they are kept and how they are deleted. Work through these questions with your legal team. Our KVKK generative AI guide sums up what the Turkish data protection authority's guide asks of companies. We cover the choice between a cloud model and a local model under KVKK in our local LLM vs ChatGPT comparison.
Hermes Agent vs OpenClaw: how do you migrate?
#You migrate from OpenClaw to Hermes Agent with Nous Research's official agent-importers plugin. According to the Migrate from OpenClaw guide, after installing the plugin you first run a preview and then start the import:
hermes plugins install agent-importers
hermes import-agent openclaw --dry-run # preview only
hermes import-agent openclaw # preview, confirm and import
On a fresh install you may not need these commands, because hermes setup offers the import by itself when it finds an ~/.openclaw folder. According to the guide, the import brings over:
- The persona file (
SOUL.md), long-term memory and the user profile - Skills from four different locations, into
~/.hermes/skills/openclaw-imports/ - The default model and custom provider settings
- MCP servers and messaging platform settings
The two routes handle API keys differently. The manual command moves keys only when you add the --migrate-secrets option, while an import through hermes setup adds that option for you. Before the import starts, a zip backup is saved under ~/.hermes/backups/. Not everything carries over: WhatsApp needs to be paired again, and OpenClaw's idle and daily session reset timers are not imported.
The same plugin is not only for OpenClaw. According to the docs' Import from Other Agents page, it also imports Claude Code, Codex CLI and Cursor settings into Hermes Agent.
Frequently asked questions
#What is Hermes Agent?
#Hermes Agent is an open-source, MIT-licensed AI agent built by Nous Research. It keeps memory across sessions, writes its own skill files after complex tasks and runs from a terminal, a desktop app or messaging apps. The user chooses which model it runs on.
Is Hermes Agent free?
#Hermes Agent is MIT-licensed and free, and the Hermes Desktop app is free too. You pay for API usage from the model provider you choose, or for an optional Nous Portal plan. As of November 12, 2026, Nous Portal has one free plan and three paid plans at $20, $100 and $200 per month.
How do you install Hermes Agent?
#On macOS and Windows the Hermes Desktop installer is recommended, and it installs the command-line tool and the desktop app together. On Linux, macOS and WSL2 the official install.sh script runs with a single curl command, and on Windows 10 and 11 the install.ps1 script runs in PowerShell. You then choose a model provider with hermes setup or hermes model.
Which Windows and Mac versions does Hermes Desktop support?
#The Windows desktop package needs Windows 11 22H2 or later, and on Windows 10 you can install Hermes Agent as a command-line tool with PowerShell. On a Mac, the ready-made desktop package is for Apple Silicon only. On Intel Macs you can install the command-line version and open the desktop app with the hermes desktop command.
Does Hermes Agent work with a local model?
#Yes. In the hermes model command you choose Custom endpoint and enter the address of Ollama, vLLM, llama.cpp, SGLang or another OpenAI-compatible server. The model's context window must be at least 64,000 tokens, and in Ollama that means raising num_ctx or the OLLAMA_CONTEXT_LENGTH variable.
What are Hermes Agent Skills?
#A skill is a SKILL.md file that walks Hermes through a specific task step by step, and the agent loads its full content only when a task needs it. Bundled skills come with the install, more can be added from the Skills Hub, and the agent can write its own skill after complex tasks. A skill installed from outside changes how the agent behaves, so review it like code before installing.
Does Hermes Agent collect data?
#According to the official FAQ, Hermes Agent collects no telemetry or analytics, and API calls go only to the provider you configure. Conversations, memory and skills stay on your computer in the ~/.hermes/ folder. If you use a cloud model or a messaging platform, the relevant part of the content goes to those services as well.
Is migrating from OpenClaw to Hermes officially supported?
#Yes. The hermes import-agent openclaw command in Nous Research's official agent-importers plugin imports the persona file, memory, skills, model settings, MCP servers and messaging settings. You can preview first with --dry-run, and with the manual command API keys move only with the --migrate-secrets option.
Sources
#- Nous Research, Hermes Agent homepage: features, Windows download options, Nous Portal plans, prices and usage multiples
- Nous Research, hermes-agent GitHub repository: the README's description, model providers, interfaces, gateway platforms, cron, subagents and MCP
- Nous Research, hermes-agent LICENSE: the MIT license and copyright line
- Nous Research, Hermes Agent v0.23.1 release page: the current release and its release date
- Hermes Agent docs, Quickstart: the recommended install path, install sequence, setup modes, OpenAI Codex subscription, 64,000-token minimum, Docker advice and hermes doctor
- Hermes Agent docs, Installation: desktop packages, the Apple Silicon requirement, .appinstaller, the Git, curl, tar and SHA-256 prerequisites and Git setup on Windows
- Hermes Agent docs, Platform Support: the Windows 11 22H2 requirement, the PowerShell install on Windows 10 and 11, hermes desktop on Intel Macs and Termux
- Hermes Agent docs, FAQ & Troubleshooting: pricing, the sudo warning, the Custom endpoint example, memory versus skills, telemetry and local storage
- Hermes Agent docs, Skills System: SKILL.md, progressive loading, the Skills Hub, the security scan, agentskills.io, and modifying and deleting skills
- Nous Research, Hermes Desktop page: the desktop app being free
- Hermes Agent docs, Hermes Desktop docs: the shared agent core and the hermes desktop command
- Hermes Agent docs, Messaging Gateway: the single process, the setup command and the allowlist default
- Hermes Agent docs, Migrate from OpenClaw: migration commands, what is imported, API keys, the backup and the settings that are not imported
- Hermes Agent docs, Import from Other Agents: the agent-importers plugin and the other supported agents
- Ollama, Ollama FAQ: the default context window, the num_ctx parameter and the OLLAMA_CONTEXT_LENGTH variable
