As of 18 November 2026. All three projects ship new releases several times a week, so details here will age quickly. Every fact comes from the projects' own pages and repositories, checked on that date and listed at the end; "Not stated" means we could not find it there.
If you want a personal agent you can run yourself, three open-source projects stand out: Nod from n8n Labs, Hermes Agent from Nous Research and OpenClaw from the OpenClaw Foundation. All three run on hardware you control and let you choose the model. They differ in who stands behind them, how much setup they expect and how much safety you get without configuring anything.
Which Hermes?
Nous Research is known for Hermes, a family of open language models. Hermes Agent is something else: an open-source agent harness under the MIT licence that runs on any model — Nous Research's or anyone else's. In this article, "Hermes" always means Hermes Agent. Its official home is hermes-agent.nousresearch.com and the NousResearch organisation on GitHub; we only used those.
Three agents in brief
Nod (n8n Labs)
A desktop app for macOS 14 or later, with Windows 11 and Linux in preview, or a self-hosted server for one person. Apache-2.0, free, and a research preview since 10 November 2026. It connects to more than 30 model providers or a local model, and is built around approvals: a separate check before it sends, pays, posts or deletes, and a Vault so the model never sees your passwords. Each installation has one owner, and Nod does not need n8n itself.
Hermes Agent (Nous Research)
MIT-licensed, and pitched as an agent with a built-in learning loop: it curates its own memory, searches its past sessions, builds a model of you over time, and creates skills after complex tasks and improves them in use, in the open agentskills.io format. It works with Nous Portal, OpenRouter, OpenAI, your own endpoint and many other providers, and switching is one command. It runs on Linux, macOS, WSL2, native Windows and Android via Termux, and its README lists six terminal backends — local, Docker, SSH, Modal, Daytona and Vercel Sandbox — so it can live anywhere from a $5 VPS to a GPU cluster or serverless infrastructure. One gateway serves Telegram, Discord, Slack, WhatsApp, Signal, email and the command line. Nous Portal, an optional subscription, bundles models and tools behind one sign-in.
OpenClaw (OpenClaw Foundation)
MIT-licensed, "an open-source AI assistant that runs on your own computer". It began as Clawdbot in November 2025, built by Peter Steinberger and the community, became Moltbot on 27 January 2026 and OpenClaw on 30 January 2026. It is now stewarded by the OpenClaw Foundation, an independent 501(c)(3) that employs the core team and signs releases; the project says OpenAI is a donor, not an owner. There is no paid tier, hosted service or token. One Gateway runs it on a laptop or as a shared team deployment, with native apps for macOS, iOS, Android, Windows and Linux and more than 25 chat channels, from Discord, iMessage and Teams to Google Chat and Signal.
Side by side
| Criterion | Nod | Hermes Agent | OpenClaw |
|---|---|---|---|
| Licence and source | Open source, Apache-2.0 | Open source, MIT | Open source, MIT |
| Governance | n8n Labs, a team inside the company n8n | Nous Research, an AI lab | OpenClaw Foundation, an independent 501(c)(3) |
| Who picks the model | You: 30+ providers or a local model | You: Nous Portal, OpenRouter, OpenAI, your own endpoint and others | You: hosted or local providers, as swappable plugins |
| Where the agent runs | Your computer, or your own server for one person | Your machine, a VPS, Docker, SSH or serverless backends (Modal, Daytona, Vercel Sandbox) | One Gateway on a laptop or as a shared team deployment; Docker, Nix and other paths |
| Status | Research preview since 10 Nov 2026; v0.8.4 on 17 Nov 2026 | v0.23.2 on 16 Nov 2026 | v2026.11.5 on 17 Nov 2026 |
| Criterion | Nod | Hermes Agent | OpenClaw |
|---|---|---|---|
| Availability in the EU and Switzerland | Wherever you install it; your model provider's terms apply | Wherever you install it; your model provider's terms apply | Wherever you install it; your model provider's terms apply |
| Where memory and credentials live | On the machine running Nod; secrets in the Vault, never seen by the model | On the system you run it on | On your hardware |
| Training use | Set by your model provider | Set by your model provider | Set by your model provider |
| Telemetry defaults | Usage analytics on by default; conversation traces off | Not stated in the README | A daily version check; anonymous feature statistics opt-in; both can be turned off |
| Approval and safety model | A separate check before it sends, pays, posts or deletes; rules enforced by a server; actions logged | Command approval, DM pairing, container isolation | Inbound messages treated as untrusted; unknown DM senders must be paired |
| Default security posture | Sandboxed computer and browser; secrets hidden from the model; asks when unsure | Not stated; depends on the backend you choose | Tools run on the host for the main session unless you configure sandboxing |
| Criterion | Nod | Hermes Agent | OpenClaw |
|---|---|---|---|
| Integrations and channels | Gmail, Calendar, Drive, Slack, websites, MCP, APIs; app, Telegram, WhatsApp | Telegram, Discord, Slack, WhatsApp, Signal, email, CLI; skills in the agentskills.io format | Discord, iMessage, Slack, Teams, Telegram, WhatsApp and 20+ more; tools, skills, plugins |
| Install and running effort | Desktop installer; about ten minutes by n8n's estimate | One-line installer, then a CLI and a messaging gateway to set up | Installer script or npm, then an onboarding wizard; security guides to read before exposing it |
| Pricing model | Free; you pay your model provider | Free; optional Nous Portal subscription for models and tools | Free; no paid tier, hosted service or token |
| Who it suits | One person who wants a guided app with approvals | Technical users who want an agent that learns, on any backend | Users who want the most channels, a foundation-run project or a shared team gateway |
Where each one wins
- Nod: the guided app and the approval layer. It is the most finished experience for one person — installer, onboarding, a permission check enforced by a server, the Vault, per-task receipts and a spending limit, out of the box. The price is a research preview, one owner per installation and usage analytics on by default.
- Hermes Agent: the skills and learning loop, and any backend. It turns experience into skills in an open, portable format, and moves freely between your laptop, a VPS, containers and serverless sandboxes that cost almost nothing while idle. The price is more operating work, and a README that does not document telemetry.
- OpenClaw: the most channels, foundation governance and the largest community of the three on GitHub. A foundation that employs the core team is a different bet from a company's Labs experiment, and one Gateway can serve a whole team. The price is security you must configure: tools run on the host unless you set up sandboxing.
Running and securing each
Nod is the shortest path. Put it on a machine that stays awake if you want schedules, leave approvals on, keep the spending limit and decide whether usage analytics stay on. Choose the model deliberately: an OpenAI project in the EU region, Mistral, or a local model.
Hermes Agent asks you to choose a backend, and that choice is your isolation boundary: a container or a remote sandbox keeps commands away from your own files better than the local backend does. Use command approval, and pair messaging users before they can talk to it.
OpenClaw's own guidance is the right starting point: read the security guide, the exposure runbook and the sandboxing guide before connecting other users or exposing the Gateway. Keep DM pairing on, treat every inbound message as untrusted, and configure sandboxing rather than letting tools run on the host.
Two things apply to all three. Whichever agent you pick, the model provider sees what each task needs, so the model decides your data's real destination. And all three release several times a week, so updating is part of running them. If you want one of them on a server you control, set up and hardened, tell us what needs solving.
Sources (checked 18 November 2026)
- Nod — official site
- Nod source code and releases on GitHub
- Hermes Agent — Nous Research
- Hermes Agent on GitHub
- OpenClaw on GitHub
- OpenClaw releases
- OpenClaw documentation
- OpenClaw lore — the project's name history
Frequently asked questions
- Is Hermes Agent the same as the Hermes models?
- No. Nous Research also publishes the Hermes family of language models. Hermes Agent is a separate open-source agent harness under the MIT licence, and it can run on any model, not only Nous Research's own.
- Which of the three is easiest to set up?
- Nod, by design: a desktop installer and, by n8n's own estimate, about ten minutes to a first task. Hermes Agent and OpenClaw install with a one-line script or a package manager and expect more configuration, especially for messaging channels and sandboxing.
- Do any of them cost money?
- None of the three charges for the software. You pay the model provider you connect, or run a local model. Hermes Agent offers an optional Nous Portal subscription that bundles models and tools under one sign-in.
- Are they safe to connect to my email and accounts?
- Each has safety features, and none is a guarantee. Nod checks each send, payment, post or deletion against your request; Hermes Agent offers command approval, DM pairing and container isolation; OpenClaw pairs unknown senders and asks you to read its security guides before exposing it. Start with read-only tasks and keep approvals on.
- Can I keep the data in Europe?
- You can run all three on a server in a European data centre and connect a European or local model. The model provider you choose still receives what each task needs, so that choice matters as much as where the agent runs.