Side-by-side
| Axis | OpenClaw | Hermes Agent |
|---|---|---|
| Setup | 10-15 min on 2026.4. Significantly more involved than the original one-line install. | One script (install.sh) on Linux, macOS and WSL2, a PowerShell one-liner on native Windows, then hermes setup. Docker images exist if you prefer a container. |
| Security model | Sandbox off by default: the main session's tools run on the host until you configure one (Docker, Podman, SSH and others). Its gateway takes a token. Governed by the OpenClaw Foundation, a US non-profit, since July 2026. | Commands run on the host by default (local backend); those matching its dangerous-pattern list need approval, smart mode by default. Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox backends are opt-in. The chat gateway refuses everyone until allowlisted or paired. |
| Model support | Multi-LLM: Claude, GPT, Gemini, Mistral, OpenRouter, local via Ollama. | Nous Portal, OpenRouter, Anthropic, OpenAI, Gemini and local servers such as Ollama and vLLM, per its providers docs. |
| Cost | $5/mo VPS works. AI API costs separate. Plugin marketplace mostly free. | Free software (MIT); its README says it runs on a $5 VPS. The installer adds a browser tool with its own Chromium unless you pass --skip-browser. Model costs separate. |
| Ecosystem | Largest plugin ecosystem in the category. Most third-party tools target OpenClaw first. | About 255,000 GitHub stars (October 2026), a plugin directory, and skills that follow the agentskills.io standard. hermes claw migrate imports an OpenClaw setup. |
| Best for | If you have an existing OpenClaw install with custom plugins or team familiarity. | If you're starting fresh today and don't have a specific reason to choose otherwise. |
Verdict
For new deployments, Hermes Agent is the easier on-ramp. OpenClaw 2026.4+ is fine: it's just no longer the obvious default. Neither sandboxes tool execution by default: switch it on whichever you pick.
Notes
- OpenClaw's post-crisis stabilisation is real but new users hit edge cases more often than Hermes users do.
- Hermes' security policy treats its approval gate as a heuristic: for production or untrusted input it asks for the whole agent inside its Docker image or NVIDIA OpenShell.
- Plugin ecosystem inertia favours OpenClaw for at least another quarter.
Sources
- OpenClaw README (v2026.10.4): tools run on the host unless you configure sandboxing
- OpenClaw docs: sandboxing
- Hermes Agent README (v0.22.2)
- Hermes Agent docs: tools and terminal backends
- Hermes Agent docs: security
- Hermes Agent SECURITY.md (v0.22.2): trust model
- Hermes Agent docs: AI providers
- Hermes Agent docs: migrate from OpenClaw
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.