Beginner guide · Open-source agents
OpenClaw vs Hermes Agent: a beginner's guide
Short answer: OpenClaw and Hermes Agent are both free, open-source AI agents you install on your own computer or server and chat with from apps like Telegram or WhatsApp. OpenClaw is run by a non-profit foundation and works with 25+ chat apps, including iMessage and Teams. Hermes Agent, from Nous Research, learns as it goes and writes its own skills. Both are more hands-on than an agent app, and both need careful safety settings.
Updated · By BotsDoThat
Every fact here comes from each project's official docs or GitHub README, linked inline. Software changes fast, so check the linked page before you rely on a detail.
What is OpenClaw?
OpenClaw calls itself “an open-source AI assistant that runs on your own computer and meets you in the channels you already use”: Discord, iMessage, Slack, Teams, Telegram, WhatsApp and 20+ more (OpenClaw README). A program called the Gateway runs on your machine and connects your chat apps, tools and AI model. You can use hosted models such as Anthropic, OpenAI or OpenRouter, or local models (OpenClaw FAQ: What is OpenClaw?). It's stewarded by the OpenClaw Foundation, an independent 501(c)(3), and has no paid tier (README).
Its own FAQ lists everyday uses like inbox, calendar and news briefings, research and drafting, reminders and follow-ups, and browser automation (OpenClaw FAQ).
What is Hermes Agent?
Hermes Agent is “the self-improving AI agent built by Nous Research” (Hermes Agent README). It keeps a memory across sessions, creates skills from experience and improves them as it uses them, runs scheduled jobs, and can work from Telegram, Discord, Slack, WhatsApp, Signal, email or a terminal (Hermes Agent site). You choose the model: Nous Portal, OpenRouter, OpenAI, your own endpoint and others (README). It's free under the MIT license (Hermes site).
Which “Hermes”? The Hermes people pair with OpenClaw is Nous Research's Hermes Agent: OpenClaw's docs include “Migrating from Hermes” (OpenClaw docs), and Hermes' docs include “Migrate from OpenClaw” (Hermes docs).
OpenClaw vs Hermes: side by side
| OpenClaw | Hermes Agent | |
|---|---|---|
| Who makes it | The OpenClaw Foundation, an independent 501(c)(3); built by Peter Steinberger and the community (README) | Nous Research (README) |
| Price of the software | Free and open source (MIT). “No paid tier, hosted service, or token” (README) | Free and open source (MIT). Optional paid Nous Portal subscription for models and tools; cloud hosting optional (site) |
| Setup | One-line installer for macOS, Linux and Windows, or desktop apps; needs Node 24.16+. Guided onboarding, then a web dashboard (Install, Getting started) | One-line installer for macOS, Linux and WSL2, a native Windows installer, or the Hermes desktop app for macOS, Windows and Linux (README, site) |
| Where it runs | Your own computer or a server (Mac, Linux, VPS). Minimum server: 1 vCPU, 1 GB RAM (FAQ, hosting FAQ) | Your computer or a server; tools can run locally or in Docker, SSH, Modal, Daytona or Vercel Sandbox (README) |
| Chat apps | Discord, iMessage, Slack, Teams, Telegram, WhatsApp and 20+ more (README) | Telegram, Discord, Slack, WhatsApp, Signal, email, CLI and more (site) |
| AI models | Hosted providers such as Anthropic, MiniMax, OpenAI and OpenRouter, or local models; can reuse a Claude Code or Codex login (FAQ, Getting started) | Nous Portal, OpenRouter, OpenAI, your own endpoint and others; switch with hermes model; local models via Ollama (README, Ollama guide) |
| Schedules | Automations: scheduled jobs, webhooks and Gmail triggers (Automations) | Built-in scheduler you set up in plain language (Scheduled tasks) |
| Memory | State, memory and credentials stay on your hardware (README) | Persistent memory; creates and improves its own skills from experience (README) |
| Safety controls | Gateway listens only on your own machine by default; unknown senders must pair first; openclaw security audit; sandboxing is off by default and opt-in (Security, Sandboxing) | Asks before dangerous commands (default “smart” mode); unknown users denied unless allowlisted or paired; scheduled jobs deny dangerous commands by default; optional container isolation (Security) |
| Switching | Can import a Hermes setup: openclaw migrate hermes (docs) | Can import an OpenClaw setup: hermes import-agent openclaw (docs) |
How to pick one
- Never used a terminal? Start with an agent app instead; our no-code guide walks you through one. Come back when you want to self-host.
- Want iMessage, Teams or one of OpenClaw's 25+ chat apps, and a non-profit project with no paid tier? OpenClaw.
- Want an agent that writes and improves its own skills, cloud sandboxes, or one login for models and tools (Nous Portal)? Hermes Agent.
- Not sure? Try one. Each can import the other's setup later (Hermes importer, OpenClaw importer).
What they cost to run
The software is free. You pay for the AI model and, if you don't use your own computer, a server. OpenClaw's minimum server is 1 vCPU and 1 GB of RAM (OpenClaw FAQ: providers and hosting). OpenClaw also lets you sign in with an API key or reuse a Claude or ChatGPT/Codex subscription you already have (OpenClaw FAQ: quick start). The cheapest route for Hermes is a local model through Ollama, which its docs call “Zero API Cost” (Hermes docs: Run Hermes locally with Ollama). Asked whether a local model is OK for casual chats, OpenClaw's FAQ answers “usually no” and warns that smaller models raise prompt-injection risk (OpenClaw FAQ).
For a pay-per-use model, a once-a-weekday inbox briefing comes to about $0.03 to $4.18 a month in our estimate, depending on the model, before the agent's own instructions, memory and tool lists, which add tokens to every request. Try your numbers or read how much an AI agent costs.
Safety tips for both
- Treat every incoming message as untrusted. OpenClaw's README says so directly, and both pair unknown senders before they can talk to your agent (OpenClaw, Hermes).
- Keep approvals on. In Hermes, don't switch approvals to “off” or run with
--yolo; keep scheduled jobs on “deny” (Hermes security). In OpenClaw, deny or tightly limit risky tools like running commands and the browser for any agent that reads untrusted content (OpenClaw security FAQ). - Turn on a sandbox. OpenClaw runs tools on your computer unless you configure sandboxing, which is off by default (OpenClaw docs: Sandboxing). Hermes can run tools in a Docker container instead of on your machine (Hermes security).
- Run the audit.
openclaw security audittells you if you've drifted from the safe defaults (OpenClaw security). - Keep it private. OpenClaw's Gateway only listens on your own machine by default; if you need remote access, its docs recommend authenticated private access such as an SSH tunnel or tailnet rather than opening it to the internet (OpenClaw security FAQ).
- Use a reader for email. OpenClaw's docs suggest sending untrusted mail to a read-only reader agent first (OpenClaw docs: Prompt injection); Hermes' docs show how to give the agent its own mailbox (Hermes docs: Give your agent its own email address).
The full beginner version: is it safe to give an AI agent your email and calendar?
Jobs people use them for
Any of our 12 blueprints works as instructions for either agent. Good first ones: Morning Inbox Briefing, Web Page Watcher and Scam Checker. Related: put a small team of agents on one project, each with its own job, give your agent its own email address, run open AI models on your own computer for simple jobs.
Questions people ask
What is OpenClaw?
OpenClaw is a free, open-source AI assistant that runs on your own computer or server and answers you in chat apps you already use, like WhatsApp, Telegram, Slack, Discord and iMessage. It's run by the OpenClaw Foundation, a non-profit (OpenClaw README).
What is Hermes Agent?
Hermes Agent is a free, open-source AI agent from Nous Research. It remembers what it learns, writes its own reusable skills, runs scheduled jobs, and talks to you from Telegram, Discord, Slack, WhatsApp, Signal, email or the terminal (Hermes Agent README).
OpenClaw vs Hermes: which is better for beginners?
Neither is as easy as an agent app with a connect button: both need an install and an AI model or API key. Both now offer desktop apps. Pick OpenClaw if you want chat apps like iMessage and Teams (25+ in all) and a project run by a non-profit with no paid tier. Pick Hermes if you want an agent that builds its own skills, or one login (Nous Portal) for models and tools. You can switch later: each can import the other's setup.
Is OpenClaw free?
Yes. The software is free and open source, with no paid tier or hosted service. You pay for the AI model you connect and for the machine it runs on.
Is Hermes Agent free?
Yes. Hermes Agent is free and open source under the MIT license. Model providers and optional hosted services, including Nous Portal, have their own pricing.
What's the cheapest way to run Hermes Agent?
On a computer you already own with a local model through Ollama, which Hermes' docs call “Zero API Cost” (minimum 8 GB of RAM for small 3B models; Hermes docs: Run Hermes locally with Ollama). Smaller models are easier to trick with prompt injection (OpenClaw docs: Prompt injection), so keep approvals on. Next cheapest: a small pay-per-use model; the cheapest in our price table is Groq openai/gpt-oss-20b. See what an AI agent costs.
Can I move from OpenClaw to Hermes, or back?
Yes. Hermes has an official importer for OpenClaw setups (hermes import-agent openclaw), and OpenClaw can import a Hermes setup (openclaw migrate hermes). Both preview changes first.
Is it safe to connect OpenClaw or Hermes to my email?
It can be, with care. Both docs treat incoming messages as untrusted. OpenClaw's docs suggest routing untrusted mail to a separate read-only reader agent (OpenClaw docs: Prompt injection). Keep pairing and allowlists on, keep approvals on, and give it only the accounts it needs. Our guide: is it safe to give an AI agent your email and calendar?
Sources
- OpenClaw README
- OpenClaw docs: Install
- OpenClaw FAQ: What is OpenClaw?
- OpenClaw FAQ: providers and hosting
- OpenClaw FAQ: quick start
- OpenClaw docs: Security
- OpenClaw FAQ: Security and access control
- OpenClaw docs: Prompt injection
- OpenClaw docs: Sandboxing
- OpenClaw docs: Automations
- OpenClaw docs: Migrating from Hermes
- Hermes Agent README
- Hermes Agent site
- Hermes docs: Security
- Hermes docs: Run Hermes locally with Ollama
- Hermes docs: Scheduled tasks
- Hermes docs: Migrate from OpenClaw
- Hermes docs: Give your agent its own email address
- Hermes docs: Tips and best practices