What Is Hermes Agent? Nous Research's Self-Improving AI Agent Explained
What Hermes Agent from Nous Research is, how its self-written skills work, what self-hosting it takes, what Nous Portal costs, and how it compares with OpenClaw.
At a glance
| Bot | License | Chat apps | |||||
|---|---|---|---|---|---|---|---|
| Hermes Agent | MIT | 261.7K | Yes | No | Telegram, WhatsApp, iMessage, Discord, Slack | From $20/mo | |
| OpenClaw | MIT | 394.2K | Yes | No | Telegram, WhatsApp, iMessage, Discord, Slack | Free | |
| NemoClaw | Apache-2.0 | 23K | Yes | Yes | Telegram, Discord, Slack | Free | |
| ZeroClaw | MIT OR Apache-2.0 | 33.3K | Yes | Yes | Telegram, WhatsApp, Discord | Free |
Checklist
| Hermes Agent | OpenClaw | NemoClaw | ZeroClaw | |
|---|---|---|---|---|
| Open-source license | Yes | Yes | Yes | Yes |
| Runs local models | Yes | Yes | Yes | Yes |
| Sandboxed by default | No | No | Yes | Yes |
| No telemetry by default | Yes | No | No | Yes |
| Lives in your chat apps | Yes | Yes | Yes | Yes |
| Works while you're away | Yes | Yes | Yes | Yes |
Hermes Agent is a free, MIT-licensed AI agent from Nous Research that you run on your own computer or server and talk to through chat apps like Telegram, Discord, Slack, WhatsApp and Signal. Nous announced it on February 25, 2026. It’s written in Python and is one of the two largest open source personal agents, alongside OpenClaw.
It is the software around a model, not a model: it works with whichever one you connect.

What makes it different: it writes its own skills
Nous calls Hermes Agent self-improving. In practice that means three things (docs):
- After a complex task, it writes a reusable skill describing how it did it, and refines that skill as it uses it again.
- It keeps a curated memory and can search across past sessions.
- It builds a profile of you over time.
The upshot is an agent that should get better at your recurring tasks without you writing prompts or plugins. It also means its behavior drifts as it learns, so it’s worth looking at what it has written into its skills now and then.
You can also run several separate agents as profiles, each with its own model, memory and skills, and put them in a group chat together.
What running it takes
Hermes Agent runs on macOS on Apple Silicon, Windows 10 or 11, or 64-bit Linux or WSL2. The installer script needs Git and curl and sets up Python 3.14 itself; macOS and Windows also have desktop packages, while Linux has no desktop package and uses the script (installation). Docker is supported too, and Android works through Termux on the command line.
You manage it from a terminal UI or a local web dashboard where you see sessions from every channel and switch chat platforms on and off. One gateway connects 20+ platforms. iMessage is the awkward one: it needs a separate BlueBubbles server running on a Mac.
Like OpenClaw, it’s built for people comfortable with a command line and with keeping a server running.
What it costs
The agent is free. You pay for the model, in one of three ways:
- Your own provider key, such as OpenRouter, OpenAI or any compatible endpoint. Pay-as-you-go.
- A local model, free per message (local models guide).
- Nous Portal, Nous’s subscription: Plus at $20 a month, Super at $100 and Ultra at $200, with 200+ models, hosted web search, image generation, speech and a cloud browser, plus one hosted cloud agent (4 to 8 GB RAM depending on tier). Portal also has free models.
Hosted web search, images and speech need either a Portal plan or your own keys for those services.
Privacy and safety defaults
Hermes Agent is one of the better-behaved large agents on privacy. It stores conversations, memory and skills under ~/.hermes, sends API calls only to the model provider you configure, and its metrics for Nous are off unless you choose “Share with Nous” (FAQ).
Isolation is a different story. Its terminal backend defaults to local, which runs commands directly on your machine. When a command matches its list of dangerous patterns, the default smart approval mode has a helper model rate it: low-risk ones run, clearly dangerous ones are refused, and you’re asked about the rest (security docs). To get real isolation, set terminal.backend: docker. Chat access is denied to everyone until you set allowlists or approve people through DM pairing. Details in Is OpenClaw safe?, which covers Hermes too.
Hermes Agent vs OpenClaw
The two are close in what they do. The differences that matter for self-hosting, from the table above:
- Telemetry: Hermes sends none by default; OpenClaw runs a daily update check that reports version and system details.
- Language: Hermes is Python, OpenClaw is TypeScript on Node.
- Business model: OpenClaw is run by a nonprofit foundation with nothing to sell; Nous sells Portal subscriptions next to the free code.
- Ecosystem: OpenClaw has more channels, providers and plugins.
- Mobile: OpenClaw has iOS and Android apps; Hermes has no native mobile app.
Hermes can import OpenClaw’s settings, memories and skills in one command, so switching is cheap. For the full comparison, see Hermes Agent vs OpenClaw and what OpenClaw is.
Related projects
- NemoClaw, NVIDIA’s alpha stack, can run Hermes Agent inside a locked-down OpenShell sandbox with egress control. See what NemoClaw is.
- ZeroClaw is a smaller alternative that ships with an OS sandbox on and no telemetry.
More options in Hermes Agent alternatives.
Features
| Hermes Agent | OpenClaw | NemoClaw | ZeroClaw | |
|---|---|---|---|---|
| Web search | Yes | Yes | Yes | Yes |
| Image generation | Yes | Yes | Unknown | Yes |
| Voice mode | Yes | Yes | Unknown | Yes |
| File uploads | Yes | Yes | Unknown | Yes |
| Memory | Yes | Yes | Unknown | Yes |
| Takes actions for you | Yes | Yes | Yes | Yes |
| App integrations | Yes | Yes | Yes | Yes |
| Works on its own | Yes | Yes | Yes | Yes |
| Messages you first | Yes | Yes | Unknown | Yes |
| Runs several bots | Yes | Yes | Unknown | Yes |
Sources
Hermes Agent
- NousResearch/hermes-agent on GitHub (MIT, v0.23.0)
- Hermes Agent homepage and Nous Portal pricing
- Hermes Agent documentation
- Hermes Agent docs: Profiles, running multiple agents
- Hermes Agent docs: Bot Mode
- Hermes Agent docs: Voice mode
- Hermes Agent docs: Vision and image paste
- Hermes Agent docs: iMessage via BlueBubbles
- Nous Research on X: Meet Hermes Agent (February 25, 2026)
- Hermes Agent docs: Session heartbeats and cron watchdogs
- Hermes Agent docs: FAQ (Is my data sent anywhere?)
- Hermes Agent docs: Installation (desktop packages, install script, Docker)
- Hermes Agent docs: Platform support (no Linux desktop package; PyPI installs unsupported)
- Hermes Agent docs: Docker setup
- Hermes Agent docs: Configuration (terminal backend defaults to local)
- Hermes Agent docs: Security (approval modes, optional container isolation)
- Hermes Agent docs: Local models
- Hermes Agent example config (shared metrics off by default)
OpenClaw
- OpenClaw homepage and Foundation section
- OpenClaw integrations: channels, providers and plugins
- openclaw/openclaw on GitHub (MIT, v2026.11.1)
- OpenClaw docs: Sandboxing (off by default)
- Wikipedia: OpenClaw
- Notebookcheck: OpenClaw can burn through hundreds of dollars per day
- OpenClaw docs: Memory overview
- OpenClaw docs: Automations
- OpenClaw docs: Chat channels
- OpenClaw docs: iMessage
- OpenClaw docs: Heartbeat, now a default Proactive check job
- OpenClaw docs: Multi-agent routing
- OpenClaw docs: Platforms (macOS, Windows, Linux, iOS and Android apps)
- OpenClaw docs: Tools (exec, browser, web search)
- OpenClaw docs: Audio and voice notes (inbound transcription)
- OpenClaw docs: Image and media support (inbound media)
- OpenClaw docs: Install (system requirements, installer script, npm, desktop apps, Docker)
- OpenClaw docs: One-click deploy on Railway
- OpenClaw docs: Usage telemetry and update checks (on by default, feature statistics opt-in)
- OpenClaw docs: Ollama provider (local models)
NemoClaw
- NVIDIA/NemoClaw on GitHub (Apache-2.0, README: supported agents, alpha status)
- NemoClaw docs: Overview
- NemoClaw docs: Prerequisites and platforms
- NemoClaw docs: Choose an inference provider (Ollama, vLLM, NIM)
- NemoClaw docs index: messaging channels (WhatsApp, Teams, Google Chat, WeChat experimental)
- NemoClaw source: telemetry delivery and NEMOCLAW_DISABLE_TELEMETRY
ZeroClaw
- zeroclaw-labs/zeroclaw on GitHub (MIT OR Apache-2.0, v0.8.7, README)
- ZeroClaw website
- ZeroClaw docs: Sandboxing
- ZeroClaw source: risk presets (Balanced, recommended, has the sandbox on)
- ZeroClaw docs: Channels overview (default and standard release bundles)
- ZeroClaw docs: You own it (no telemetry)
- ZeroClaw docs: Docker and containers
- ZeroClaw source: tools (image generation, screenshot, web)



