What is Hermes Agent? Features, uses and comparisons

Who makes Hermes Agent and what does it do?

Hermes Agent is an open-source AI agent from Nous Research that runs on hardware you control and becomes more useful the longer you use it. The source code is on GitHub at NousResearch/hermes-agent under the MIT license, and the official documentation lives at hermes-agent.nousresearch.com/docs.

Nous Research describes it as a self-improving agent with a built-in learning loop. In practice that means three things: it writes reusable skills after it works out a hard task, it keeps a small curated memory about you and your environment, and it can search its own past conversations.

The docs position Hermes as an autonomous agent rather than a coding copilot tied to an IDE or a chat wrapper around a single API. It is not tied to your laptop. You can run it on a small VPS, a GPU cluster, or serverless infrastructure such as Daytona or Modal, and talk to it from Telegram while it works on a remote machine.

What are the main features of Hermes Agent?

Hermes Agent combines a terminal interface, a messaging gateway, persistent memory, a skills system, a scheduler, subagents, several execution backends and MCP support in one install. The table summarizes each part.

Feature	What it does
Learning loop	Agent-curated memory with periodic nudges, skill creation after complex tasks, skills that improve during use, and full-text search over past sessions with LLM summaries
Skills	On-demand `SKILL.md` instruction files that follow the agentskills.io open standard. Every installed skill becomes a slash command
Memory	Two small files, `MEMORY.md` (the agent's notes) and `USER.md` (your profile), loaded into the system prompt when a session starts
Messaging gateway	One gateway process connects the agent to Telegram, Discord, Slack, WhatsApp, Signal, email and more. The docs list over 20 platforms
Cron scheduler	One-shot or recurring jobs written in natural language or cron syntax, with results delivered to a chat, a file or another platform
Subagents	The `delegate_task` tool spawns child agents with isolated context. Up to 10 run in parallel by default
Terminal backends	Seven places to run commands: local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox
MCP	Connect Model Context Protocol servers through `mcp_servers` in `config.yaml`, with per-server tool filtering

A few other parts matter once you go beyond one agent. Profiles let you run several independent agents on one machine, each with its own config, memory and bot tokens. The kanban board lets those profiles hand work to each other. Hermes Desktop is a desktop app, with installer packages for macOS and Windows, that can also connect to Hermes backends on other machines.

You choose the model. Hermes works with Nous Portal, OpenRouter, OpenAI, Anthropic, many other providers and any OpenAI-compatible endpoint, and you switch between them with hermes model. Whatever you pick needs a context window of at least 64,000 tokens.

How does the learning loop work?

The learning loop is the combination of memory, skills and session search, and each part holds a different kind of knowledge.

Memory holds small facts that should always be in context. MEMORY.md is capped at 2,200 characters and USER.md at 1,375 characters. The agent adds, replaces and removes entries itself with the memory tool.
Skills hold longer procedures that load only when a task needs them. The agent saves a skill with the skill_manage tool when it worked out a multi-step workflow, found a working path after errors, or was corrected by you.
Session search fills the gaps. When an old conversation has left the context window, the agent can search past sessions and get summaries back.

Memory is a frozen snapshot taken when a session starts, so a fact saved mid-session appears in the next session. On messaging platforms, where one chat can run for weeks as a single session, the docs recommend sending /new at natural boundaries such as a finished task or a change of topic.

If you want to review what the agent learns before it sticks, turn on write approval for memory and for skills. Changes then wait for your approval.

How does Hermes Agent compare with OpenClaw, Claude Code and Codex?

Hermes Agent shares several file conventions with OpenClaw and ships import commands for OpenClaw, Claude Code and the OpenAI Codex CLI, so you can bring an existing setup across. The table sticks to what the Hermes docs state about each tool. "Not covered" means the Hermes docs do not describe that part.

	Hermes Agent	OpenClaw	Claude Code / Codex CLI
Persona file	`SOUL.md` in the Hermes home	`workspace/SOUL.md`	Not covered
Instruction files	`.hermes.md` or `AGENTS.md` per project, with `CLAUDE.md` and `.cursorrules` also recognized	`workspace/AGENTS.md`	`CLAUDE.md` (Claude Code), `AGENTS.md` (Codex CLI)
Memory	`MEMORY.md` and `USER.md`, managed by the agent	`workspace/MEMORY.md`, `USER.md` and daily memory files	Codex `memories/*.md` files
Skills	`SKILL.md` folders the agent can create and edit	Workspace and shared skill folders	`skills/<name>/` folders with a `SKILL.md`
MCP servers	`mcp_servers` in `config.yaml`	`mcp.servers.*`	`mcpServers` (Claude Code), `[mcp_servers.*]` (Codex CLI)
Import into Hermes	Not applicable	`hermes claw migrate`	`hermes import-agent`

Both import commands show a preview before they write anything. hermes claw migrate brings API keys across only when you add --migrate-secrets, and hermes import-agent never reads the Claude Code or Codex credential files.

The main difference in scope, according to the docs, is that Hermes is built to run as a long-lived agent with a messaging gateway, a scheduler, profiles and remote execution backends. It can still sit alongside coding tools: it can use a ChatGPT or Codex subscription as its model provider, and hermes acp connects it to editors through ACP.

Who is Hermes Agent for?

Hermes Agent suits people who want an agent that stays online, remembers context and runs on infrastructure they control. Setups the docs describe include:

A personal assistant on one Telegram bot and a coding agent on another
One agent per family member or per Slack workspace
A research agent, a writing agent and a cron-driven bot, each with separate memory and skills
Recurring daily briefs and scheduled reports delivered to a chat
Engineering pipelines that split work across several named agents on a kanban board

How do you get started with Hermes Agent?

You can install Hermes Agent with one command and have a working chat a few minutes later. On Linux, macOS or WSL2:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
source ~/.bashrc    # or: source ~/.zshrc
hermes setup
hermes

On native Windows, run this in PowerShell:

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

If you prefer a desktop app on macOS or Windows, download the Hermes Desktop installer from the Hermes website instead. If you would rather not collect separate API keys, hermes setup --portal signs you in to Nous Portal and sets up a model plus the Tool Gateway in one step.

A sensible order after installing:

Get one plain chat working with hermes setup or hermes model.
Edit ~/.hermes/SOUL.md to set the agent's voice.
Connect a messaging platform with hermes gateway setup.
Add skills, cron jobs and MCP servers once the basics work.
Run hermes doctor when something looks wrong, and hermes update to stay current.

Is this site part of Nous Research?

No. This site is an independent directory of guides and community projects for the Hermes Agent ecosystem, and it is not affiliated with or endorsed by Nous Research. For authoritative details, check the official documentation and the GitHub repository, which are the sources for this guide.

FAQ

Who makes Hermes Agent?

Nous Research builds Hermes Agent. It is open source under the MIT license and developed in the NousResearch/hermes-agent repository on GitHub.

Is Hermes Agent free?

The software is free and open source under the MIT license. You still pay for the model you use, whether through a provider API key, a subscription such as Nous Portal, or hardware for a local model.

Which models does Hermes Agent support?

Hermes works with Nous Portal, OpenRouter, OpenAI, Anthropic, Google, many other providers and any OpenAI-compatible endpoint. The model needs at least 64,000 tokens of context, and you switch models with hermes model.

Can I move from OpenClaw to Hermes Agent?

Yes. Run hermes claw migrate to import your SOUL.md, memories, skills, command allowlist and messaging settings, and add --migrate-secrets if you also want API keys. Use --dry-run first to preview the changes.

Where can Hermes Agent run?

It installs on Linux, macOS, WSL2 and native Windows, and there is a separate Termux package for aarch64 Android devices. Its commands can run locally or on Docker, SSH, Singularity, Modal, Daytona or Vercel Sandbox backends.

What is Hermes Agent?