OpenClaw is an AI agent that actually operates things — browsing the web, running tasks, connecting to messaging platforms. That means its architectural considerations are completely different from those of a plain chatbot.
The 30-second overview
| Item | Details |
|---|---|
| Deployment | An isolated Docker container on an isolated network |
| Where the API key lives | Inside your container, AES-256 encrypted |
| Browser automation | The main source of resource usage |
| Why isolation matters | It runs tasks and it stores credentials |
| How the plans differ | Both include browser automation; the difference is storage and AI — Pro has 10 GB, Advanced has 20 GB plus Roamer AI |
Your API key lives in your own container
On the Pro plan you bring your own AI model API key (Advanced answers with Roamer AI by default, so no key is needed, though you can still add your own). That key is written only into your own container and is never stored centrally in the platform’s database.
What this design means in practice: you pay model call costs directly to the model provider, and usage, which model to use, and whether to configure several at once are all your decisions. You can also switch providers at any time without waiting for platform support.
How the two plans differ
| Plan | Monthly fee | CPU | Storage | Browser automation | Roamer AI |
|---|---|---|---|---|---|
| Pro | NT$599 | 2 cores | 10 GB | Yes | Not included |
| Advanced | NT$1,599 | 2 cores | 20 GB | Yes | Commercial access |
Both plans include browser automation, and both are provisioned to run a headless browser reliably. That is what browser automation actually needs — running a headless browser (loading pages, executing JavaScript, rendering the DOM) is resource-hungry on its own, and current versions of OpenClaw need more resources just to stay running, which is why the old Starter plan has been discontinued.
So whether you only do conversations, API calls and text processing, or want it to scrape sites, fill in forms and click buttons, Pro is enough to start with.
Advanced adds two things: double the storage (20 GB, versus 10 GB on Pro) and commercial Roamer AI access — RoamerHost’s AI engineer works directly inside your OpenClaw, creating agents, writing their personality (SOUL) and working rules, installing skills, setting up schedules and reading logs to debug. The number of custom agents differs too: 1–5 on Pro, 5–10 on Advanced.
Restart vs rebuild from scratch
There are two easily confused operations in the dashboard:
- Restart: stops the container and starts it again, and your data is fully preserved. Use this for connection problems, a dashboard that will not open, or CORS settings gone wrong.
- Rebuild from scratch: deletes the container and recreates it from the latest image, and data inside the container is wiped. Use it to upgrade versions, or when the configuration is broken beyond repair.
The difference is irreversible, so before you rebuild, make sure the data you need has been backed up. Detailed steps are in the rebuild guide.
Container-level security settings
Because OpenClaw runs tasks and may handle untrusted web content, container-level isolation matters especially. Every instance’s container has: its own Docker network (no connectivity to other instances), no-new-privileges (no privilege escalation), and hard memory and CPU limits.
How it differs from a typical SaaS
With a SaaS-style AI assistant, your conversations and data live in their system, sharing one environment with every other user.
OpenClaw is deployed differently: one subscription, one container — your API key, conversations, skills and workspace files all sit in your own environment.
That difference matters substantively in two situations: the content you handle has confidentiality requirements, or the services you connect to would rather their credentials were not sitting next to someone else’s. Otherwise it is more peace of mind than a functional difference. Better to say so honestly.
Why browsers are so resource-hungry
A browser does more than most people think: parse HTML, run JavaScript, render the layout, load images and fonts. That is the same workload as opening a web page on your own computer.
The difference is that your computer has 16 GB of memory, and the container has only what the plan gives it.
So both plans have enough resources for browser automation, and the choice comes down to how much storage you need (10 GB or 20 GB), and whether you want AI to build your agents for you.
| Use case | Recommended plan |
|---|---|
| Mostly text conversations and API calls | Pro (NT$599) |
| You need to actually run a browser | Pro (NT$599) and up |
| Heavy data collection, long-running tasks | Pro (NT$599) and up; if the data you keep needs more room, pick Advanced (20 GB) |
If you need more storage, or want AI that builds and tunes your agents for you (creating agents, writing rules, setting schedules), pick Advanced (NT$1,599): double the storage at 20 GB, plus commercial Roamer AI access.
Why isolation matters
OpenClaw is not like an ordinary tool: it runs tasks, it stores your API keys, and it may operate your accounts.
So every instance is an isolated container on an isolated network, with no shared runtime. This is not only a performance consideration — it means “your credentials are not running in the same place as someone else’s”.
FAQ
Q: Is the API key really safe?
It is stored inside your own container, AES-256 encrypted. But safety is relative — we still recommend keeping a copy in a password manager, and reviewing periodically for unused keys that should be revoked.
Q: Do the plans differ in features?
The core features (chat, API calls, scheduling, skills, messaging integrations) and browser automation are the same on both. Where they differ is storage and AI: Advanced has 20 GB of storage (10 GB on Pro) and adds commercial Roamer AI access — AI that builds agents, writes rules, sets schedules and debugs for you. The number of custom agents differs too (1–5 on Pro, 5–10 on Advanced).
Q: Does restarting the container lose data?
Restarting does not; rebuilding from scratch does. The difference is covered in the rebuild guide.
Q: Can I install it on my own VPS?
You can, but you handle container orchestration, the browser environment, credential management and updates yourself. Setting up the browser automation environment is the most troublesome part of that list.
When to upgrade your plan
Upgrading to Advanced gets you double the storage (20 GB) plus Roamer AI. The signal to upgrade is running short on storage, wanting AI to build agents, write rules, set schedules and debug for you, or needing more than 5 custom agents.
As far as resources go, only two signals mean you are genuinely short on memory, and the fix is to run fewer browser tasks or agents at once:
- Tasks stall halfway through — browser automation is out of memory
- Complex pages regularly fail to finish loading — same cause
Most other symptoms have nothing to do with resources: slow replies are the model’s response time, wrong actions mean the instructions were unclear, and picking the wrong tool means the description was too vague. An upgrade improves none of those three.
What a resource shortage actually looks like
| Symptom | Usually |
|---|---|
| Tasks stall halfway through | Out of memory, the browser gets killed |
| Complex pages never finish loading | Same cause |
| Even simple text tasks are slow | Not resources — that is model response time |
| Sometimes fails, sometimes succeeds | Could be resources at the edge, or an unstable web page |
The last row is the hardest to judge. Run the test five times in a row; if all five succeed, it is not a resource problem.
Plans can be changed at any time
If you are not sure which tier to pick, starting on Pro and testing is reasonable. Upgrades take effect immediately, with no rebuild and no impact on your data.
Run your real tasks on Pro first, and move up to Advanced once you need more storage or want AI to build your agents for you. That is far more accurate than guessing what you will need in advance.
Further reading
Want someone to build it for you?
If you would rather not assemble these flows yourself, or the scope is large enough that you want someone planning alongside you, Roamer Tech (RoamerHost’s parent company) takes on contract work in business process automation and AI agent development:
