Migrate an OpenClaw instance to the Hermes runtime — no CLI required
On Lobsterland, migrating an OpenClaw instance to the Hermes runtime is a dashboard choice, not a shell session. You pick one of two modes — Replace in place or Copy to new instance — and the managed runtime relaunches your agent on Hermes in roughly two minutes, on the same account, dashboard, and billing. No hermes claw migrate to type, no VPS to provision, no Docker or Kubernetes, no memory index to rebuild by hand.
That is the whole gap. Every other guide on this topic documents a self-hosted CLI migration: tar ~/.openclaw into ~/.hermes, stand up a second box, reindex. This is the managed version, plus the honest part nobody writes — what actually carries over, and what you lose.
- Pick a migration mode in the dashboard: Replace in place (Hermes takes over the existing instance) or Copy to new instance (a separate Hermes instance, OpenClaw stays running).
- The managed runtime does the migration as a one-shot job on your instance's own storage, then restarts. You never touch a CLI.
- Persistent memory carries across, and Hermes keeps building on it — the learning loop writes its own reusable skills over time.
- You lose the OpenClaw-only surfaces Hermes doesn't have today: WhatsApp, the workspace/file editor, cron, multiple agents, the Hosted Browser, and usage analytics.
- Hermes is bring-your-own model key (Anthropic, OpenAI, OpenRouter, Google). There's no ChatGPT/Claude OAuth bridge, so a subscription-only setup needs a plain provider key.
- Pricing doesn't change — same Light / Medium / Large plans, model usage billed by your own provider at no markup.
First, the name: Hermes here is a runtime, not a Hermes 2/3/4 model
Worth stating once plainly, because the word collides. In this context Hermes is the Nous Research open-source agent runtime — "the agent that grows with you" — not a Hermes 2, 3, or 4 language model. You're swapping the agent runtime your instance runs, the way you'd swap OpenClaw, not downloading a model.
Hermes is model-agnostic. You bring your own provider key, and through OpenRouter you can even point it at Nous's own open-weight Hermes models if you want. But the runtime and the model are separate decisions.
What carries over when you migrate from OpenClaw to Hermes
The migration runs as a one-shot job on your existing instance's storage, so your accumulated state moves with it. What carries:
- Persistent memory across sessions — and Hermes is memory-first, so it keeps compounding that context rather than treating it as scratch.
- The self-improving learning loop — Hermes writes its own reusable skills as it works, plus anything you install from Hermes Hub. This is the core reason to move: it's a different kind of agent, not a lighter OpenClaw.
- Your channels, within Hermes' set — chat in the browser, plus Telegram and Slack with the allowlist controls you'd expect (allow specific users, a whole channel, or anyone, and require an @mention in groups so it only replies when called).
- Runtime logs — you keep watching what the agent does.
- Your bring-your-own model keys — Anthropic, OpenAI, OpenRouter, and Google providers all carry.
What you lose: the OpenClaw-only surfaces Hermes doesn't have
This is the part the CLI tutorials skip, and it's the part that decides the migration. Hermes today does not offer:
- WhatsApp — Hermes channels are Telegram and Slack only.
- The workspace / file editor — OpenClaw-only.
- Cron and scheduled work — OpenClaw-only.
- Multiple agents / multi-agent routing — OpenClaw-only.
- The Hosted Browser — OpenClaw-only.
- Usage analytics — OpenClaw-only.
There's also no ChatGPT or Claude OAuth login on Hermes — it's bring-your-own API key, full stop. If your OpenClaw setup logged in with a ChatGPT or Claude subscription, you'll need a plain provider API key after migrating.
Migrate only if your workflow lives inside the Hermes surface: chat, Telegram/Slack, logs, the agent's own skills, BYO keys, and memory. If you've checked the side-by-side comparison and your day still runs on OpenClaw features, stay there.
Replace in place vs Copy to new instance: which mode to pick
There are two migration modes, and the decision is simple.
| Replace in place | Copy to new instance | |
|---|---|---|
| What it does | Flips your existing instance's runtime to Hermes, decommissions the OpenClaw setup | Stands up a separate Hermes instance from a copy of your state, leaves OpenClaw running |
| Pick it when | Hermes is taking over wholesale | You want to try Hermes side-by-side before committing |
| OpenClaw after | Gone | Still running |
If you've already decided Hermes is your agent, Replace in place is the clean move — one instance, Hermes from now on. If you're still evaluating, Copy to new instance lets you exercise Hermes against a copy of your real state while OpenClaw keeps handling production, then decommission whichever one loses.
The BYO-key reality: a provider key, not a subscription
Hermes wants a plain model-provider API key. Supported providers are Anthropic, OpenAI, OpenRouter, and Google, and via OpenRouter you can reach Nous's open-weight Hermes models.
What's not there: the ChatGPT/Claude OAuth bridge. That subscription login path isn't live for Hermes. So if your OpenClaw instance authenticated through a ChatGPT or Claude subscription rather than an API key, budget for getting a provider key before you flip the runtime — otherwise the agent will recognize the request and have nothing to answer with.
When NOT to migrate from OpenClaw to Hermes
Lead with this so you don't migrate and regret it. Stay on OpenClaw if you depend on any of:
- WhatsApp as a channel,
- scheduled / cron jobs,
- more than one agent (multi-agent routing),
- the Hosted Browser,
- the workspace / file editor,
- usage analytics, or
- a ChatGPT/Claude subscription login with no provider key to swap in.
None of those move. The whole point of an honest migration guide is that it tells you when the answer is "don't." If you're weighing the two runtimes for a team, the OpenClaw vs Hermes for teams breakdown goes deeper on the operating-model trade-offs.
How the managed migration actually runs
Under the hood it's the same restart pipeline that handles every Lobsterland instance change. The migration kicks off a one-shot init on the instance's own persistent storage that converts the OpenClaw state to Hermes, then the instance restarts onto the Hermes runtime. The instance is isolated, has no public IP, and there are no servers, Docker, or Kubernetes for you to operate. From your side it's a button and about two minutes.
That's the inversion: the rest of the field documents tarring directories and provisioning a side-by-side VPS. Here, the host runs the migration job for you.
Does the price change after migrating?
No. Hermes runs on the same plans as OpenClaw — Light $6.90/month, Medium $14.90/month, Large $29.90/month — with model usage billed directly by your own provider at no markup. Same account, same dashboard, same billing. See pricing for the full breakdown.
Migrate to Hermes, or keep evaluating
If your workflow fits the Hermes surface and you want a memory-first agent that grows with you, the migration is a two-minute, no-CLI move — and Copy to new instance lets you try it without giving up OpenClaw. If you're still weighing it, read the OpenClaw vs Hermes comparison or the Hermes runtime support announcement first.
When you're ready, start from the Hermes hub or launch directly.