Side-by-side
| Axis | Hermes Agent | Nanobot |
|---|---|---|
| Setup | One script (install.sh), then hermes model or hermes setup. Linux on x86_64 and aarch64, Apple Silicon macOS, native Windows and Docker; Android through Termux. | One script, or pip / uv / pipx install nanobot-ai on Python 3.11+. Opens a WebUI on a desktop, a terminal wizard over SSH. Wheels for macOS 13+, Linux ARM64 and x64, Windows x64. |
| Security model | Shell commands go through approvals, smart by default: a helper model approves low-risk ones, denies dangerous ones and asks you about the rest. The local backend runs on the host; the docs advise the Docker backend in production. The chat gateway refuses everyone until allowlisted or paired. | A workspace guard (restrictToWorkspace) and a shell sandbox (bubblewrap on Linux, seatbelt on macOS) exist but are off by default; the docs ask for both in production. Chat channels require pairing unless you set an allowlist (Slack and Mattermost DMs are open by default). |
| Model support | Nous Portal, OpenRouter, OpenAI, your own endpoint and many others, per its README. Switch with hermes model. | A native Anthropic provider, OpenRouter and other gateways, OpenAI-compatible APIs, and local servers such as Ollama and vLLM. |
| Cost | Free software. The README says it runs on a $5 VPS; the model bill is separate. | Free software. Model fees and hardware depend on the provider you choose, says its README. |
| Ecosystem | About 254,000 GitHub stars, 55,000 forks; v0.22.2 released on 15 October 2026. Gateway for Telegram, Discord, Slack, WhatsApp, Signal, email and more. | About 49,000 GitHub stars, 8,600 forks; v0.3.5 released on 15 September 2026. Telegram, Discord, Slack, WeChat, email, Mattermost, Feishu, Teams; MCP servers through Apps. |
| Best for | An agent that lives in chat apps on a server, with command approvals on from the first run and a choice of sandbox backends. | A smaller Python agent with a browser UI, when you are ready to turn on its workspace guard and shell sandbox yourself. |
Verdict
Both are MIT-licensed, written in Python and fine on a small Linux box with the model at a provider. The real difference is the defaults: Hermes Agent screens shell commands from the first run, Nanobot leaves its workspace guard and shell sandbox off until you set them. Pick Hermes if you will not revisit the config; pick Nanobot if you want the lighter project and will.
Notes
- Neither README gives RAM figures. Hermes Agent lists aarch64 Linux among its supported platforms (a 64-bit Raspberry Pi OS qualifies); Nanobot ships Linux ARM64 wheels.
- Nanobot's WebUI grants administration of the agent: its README keeps it on localhost and reaches it through an SSH tunnel. Do the same.
- Stars measure attention, not safety. Check each project's advisories before you deploy, and pin the version you run.
Sources
- Hermes Agent repository and README (v0.22.2)
- Hermes Agent docs: platform support
- Hermes Agent docs: security
- Nanobot repository and README (v0.3.5)
- Nanobot docs: providers and models
- Nanobot docs: configuration, security section
Going deeper
For the full landscape report including hosting economics, security posture and regulatory context, see the 2026 landscape report. For the OpenClaw-specific history, see the complete OpenClaw timeline.
New comparison requests are welcome: subscribe and reply to any edition with your short-list.