Two Agent Runtimes: OpenClaw and Hermes-AI

Update (2026-07-31): co-hosting has been removed. The /add hermes and /deploy hermes commands this post originally announced no longer exist, and no instance runs two assistants any more. Hermes-AI is still fully supported — as a selectable runtime: pick it at deploy time, or switch an existing instance with /agent. This post has been updated to match.

OpenClaw Box runs two agent runtimes. Alongside OpenClaw — the open-source assistant platform you already know — you can deploy Hermes-AI. Same 60-second Telegram deploy, same managed models, same cloud Chrome.

Two runtimes, one managed instance

An OpenClaw Box instance runs a single runtime. OpenClaw is the default and still what most people want. But a single assistant is not always the right shape — sometimes you want a different model, a different personality, or a different job.

So we added a second supported runtime: Hermes-AI (built on NousResearch's hermes-agent). You can:

• Deploy a standalone Hermes-AI instance — pick Hermes at deploy time instead of OpenClaw.
• Switch an existing instance between the two runtimes with /agent in Telegram, or from the web console.

Both options are fully managed. No servers, no Docker, no glue code.

What we tried: co-hosting

The original version of this post announced a third option: running Hermes alongside OpenClaw on the same instance, sharing one cloud Chrome, one LiteLLM model gateway, and one claude-max subscription, with a separate Telegram bot for each agent.

It worked, and the economics were appealing. But two runtimes on one instance meant two processes competing for the same browser, the same ports, and the same memory budget — and every operational question (which process owns the Telegram port, which one restarts on a config change, which one is actually answering you) had two possible answers. That ambiguity caused real outages, and made the failure modes hard to explain to the person affected.

Why we removed it

One instance, one assistant. That is the version we can operate reliably and the version whose behavior we can explain in a sentence. You still get both runtimes — you just choose which one is yours, and you can change your mind at any time:

/agent

Prefer the web console? The same switch is available there.

Who this is for

• You want a different reasoning style. Try Hermes-AI on your own workload and switch back if you prefer OpenClaw.
• You want a specialized assistant. Hermes-AI is a good fit when you want a focused agent rather than a generalist.

Get started

If you're new, deploy in 60 seconds: open @OpenClawBoxBot and type /create. Want Hermes-AI instead of OpenClaw? Pick it at deploy time in the web console, or send /agent to switch an instance you already have.

Two runtimes. Your choice.