Grok Bot vs OpenClaw: managed cloud or a framework you maintain

The short answer

Pick Grok Bot if you want agents running now, with no agent host or server to install or maintain, only the native client, and you already hold or are willing to buy an eligible Cursor or SuperGrok subscription. Pick OpenClaw if the source matters to you, if you want to change how the agent works rather than how it is configured, and if you are prepared to be the operator of a small piece of infrastructure indefinitely.

What Grok Bot actually gives you for the subscription

Grok Bot launched on 11 August 2026 in beta. You create named Bots and message them like colleagues. They work on a persistent cloud Linux VM with a browser, a filesystem and a terminal, use connectors and MCP where an API exists, and fall back to computer use, meaning driving the graphical interface, where it does not. Named Bots keep memory, files, browser sessions and preferences across turns. You can walk a Bot through a procedure once and it saves it as a routine to re-run on a schedule.

The critical detail, and the one several ranking comparison tables get exactly backwards, is that the cloud computer is not per Bot. xAI's documentation says all of your Bots use the same persistent cloud computer, that each gets its own screen on it "without getting separate security boundaries", and that you should treat a login or file placed on the computer as available to all of your Bots. If you have read a comparison table asserting a dedicated cloud computer per Bot, it is repeating the marketing line rather than the documentation.

Access, as of 22 August 2026, requires an eligible plan and a Cursor account in every case: SuperGrok Plus, SuperGrok Heavy, Cursor Pro+, Cursor Ultra, or Cursor Teams Standard or Premium. The 21 August announcement expanded it to all SuperGrok Plus, Cursor Pro+ and Cursor Teams subscribers. Cursor Ultra is $200 a month and Cursor Teams Premium is $120 per seat; SuperGrok Plus and Heavy are reported at $100 and $300, though the xAI pricing page returns 403 to us. Base SuperGrok at $30 includes Grok Build but not Grok Bot.

On top of the subscription there is an undisclosed weekly token allowance and, per eesel's teardown, no Grok Bot specific spend cap, with overage billed at raw model rates and no model picker to route work anywhere cheaper. So "managed" removes the operations work and leaves a variable bill in its place.

What OpenClaw actually costs

OpenClaw is genuinely free software, MIT-licensed, and in the alternatives lists it is consistently described as free or as "free, minus engineering time". That second phrase is accurate and almost never unpacked, so here it is unpacked.

The metered API row deserves the most attention, because it is the same risk shape Grok Bot has and it is worse in one specific way. A subscription fails safe: the window closes, the work stops, and you notice. A raw API key fails expensive: an agent looping on a task it cannot complete keeps spending until somebody looks at a dashboard. On a hosted product there is at least a vendor between you and the invoice. On a self-hosted framework driving your own keys, there is nothing between you and it unless you build it.

The maintenance tax, and the quote that names it

The strongest argument for a managed product is not features. It is that somebody else absorbs the churn. In the Hacker News thread on the Grok Bot launch, one commenter described exactly that failure mode: OpenClaw broke on every update for a month, so they gave up and moved to Hermes. One person's experience is not a verdict on a project, and fast-moving open-source software breaking during a period of rapid change is not a scandal. But it is the honest picture of what "free, minus engineering time" turns into for somebody who wanted a teammate and got a maintenance job.

The counterweight from the same thread is worth quoting too, because it points at the same underlying fact from the other side. Commenter h14h, who had been hand-rolling this setup on OpenClaw, said Grok Bot simplifies the setup about as far as they can imagine and called it a pretty slick experience. Both comments describe the same trade: the framework gives you control and hands you the operations bill; the product takes the operations bill and hands you a set of decisions you can no longer make.

Notice that the two "wins" columns barely overlap. This is a real fork in the road rather than a spectrum, which is why the alternatives lists keep placing them at opposite ends. What the lists miss is that there is a third position on this axis, and it is the one most people asking the question actually want.

The risk you take on when you become the operator

Self-hosting moves a set of risks from a vendor's balance sheet onto yours, and the alternatives lists price none of them. This is not an argument against self-hosting. It is the checklist you should work through before choosing it, and most of it applies to Grok Bot too.

• Prompt injection is not solved, on either side. The Hacker News launch thread spent most of its length on this. Commenter dgellow asked directly whether people were comfortable with agents running non-stop with access to all their accounts and getting hijacked through prompt injection, and a long subthread rejected the idea that the problem is largely handled. Commenter stymaar put the numbers argument well: a system that fails every fortieth attempt is catastrophic rather than acceptable. A self-hosted agent does not fix this; it just makes you the one who has to think about it.
• Exposure. An always-on agent you can reach from a phone is an always-on service somebody else can reach too. On a managed product the vendor owns that surface. On your VPS you own the tunnel, the auth and the patching.
• Credential blast radius. The Grok Bot criticism about a shared computer holding every login applies to a self-hosted agent as well, unless you deliberately design it not to. The default in both cases is one process with access to everything you connected.
• Dependency churn. Fast-moving open source moves fast. That is the point of it and it is also the cost of it.
• Nobody to escalate to. Grok Bot is in beta with an enterprise waitlist, which is thin support. A community project is thinner still.

The honest scoring is that Grok Bot removes the operations work and adds a vendor dependency plus a documented shared-computer design. OpenClaw removes the vendor and hands you every one of these items. Neither removes the underlying agent-safety problem, and anybody selling you either one as a solution to it is overselling.

The option both lists leave out

Read the alternatives lists closely and they present a binary: rent somebody's cloud, or become an operator. That framing quietly assumes that running an agent on hardware you own requires you to also run a server, a framework and a deployment. For coding agents specifically, it does not.

Continuum is a finished application rather than a framework. It installs on a Mac, Linux or Windows machine you already have, finds the agent CLIs already installed there, and runs Claude Code, Codex, Cursor, Grok, Gemini, and OpenCode sessions under your own logins. Each Code session gets its own git worktree so parallel agents do not collide. There is no VPS to rent, because the machine already exists and is already paid for. There is no framework to maintain, because it ships as an application with releases. And there is no local-only restriction, because the desktop app, a native iPhone client and a web client all drive the same session on the same host.

The last row is the concession, and it is a real one. Continuum is a coding workbench. It does not log into your expense portal and click through it, and it has no learn-by-demonstration routine builder for non-technical users. If that is the work, Grok Bot is the better product and OpenClaw is the better project, and you should ignore the middle column.

How to decide

Three questions, and the answers point cleanly at one of the three options above.

• Do you want to modify how the agent itself behaves? If yes, only OpenClaw can give you that, and you should accept the operations bill as the price of it. If the honest answer is that you will never open the source, the licence is not buying you anything.
• Do you own a machine that can stay powered on? If no, you are choosing between Grok Bot and a rented VPS, and the VPS route only makes sense if the first answer was yes. If yes, you can have the own-hardware property without renting anything.
• Is the work code, or is it office admin? Code points at Continuum. Office admin points at Grok Bot. A general-purpose agent you intend to build out yourself points at OpenClaw.

One further consideration for teams rather than individuals. Grok Bot defers compliance to Cursor's terms and, per eesel's review, has no published SOC 2, ISO 27001, GDPR, or HIPAA claim found in its documentation set, with an audit view documented as coming. The twenty most recent routine records are reported by Daily Dose of Data Science and unconfirmed by xAI. A self-hosted OpenClaw deployment has whatever compliance posture you give it, which is both the advantage and the catch. Neither answer is automatically the safe one, and if a security review is in your future, read the Grok Bot security and compliance page before you shortlist anything.

For the wider field, including Hermes Agent, Manus and ChatGPT Work, the alternatives guide ranks eight options on the same axes. If what you actually want to know is whether Grok Bot can run on your own machine at all, the short answer is no, and that page sets out what runs where.

Questions people ask

Keep reading

Continue through Head to head with Grok Bot vs Claude Code: an always-on teammate is not a coding agent.