Grok Bot is xAI's beta from 11 August 2026: always-on agents that each get their own computer in xAI's cloud, sign into your existing tools, and finish multi-step jobs while you are away. It is a strong product, and it validates an architecture CRHQ has been running in production since early 2026, roughly six months earlier: agents that live on an always-on machine, keep their context, and work whether or not you are watching.
The two products agree on the premise and split completely on one question. Whose machine do the agents live on? Grok Bot answers "ours, in our cloud." CRHQ answers "yours." Almost every difference below follows from that answer.
its computer
its computer
its computer
At a glance — what each platform ships out of the box
TL;DR
| Grok Bot | CRHQ | |
|---|---|---|
| What it is | Always-on agents, each with its own computer in xAI's cloud | A fleet of agents on one dedicated, managed server per customer |
| Where agents run | xAI's cloud | Your server, managed by CRHQ |
| Models | xAI's stack (no model choice published) | Claude, Codex, Grok, Gemini, routed per task; open-source providers as an option |
| Your data | Processed in xAI's cloud; governance details not published at launch | Contained on your dedicated server; you govern it, CRHQ runs security operations |
| Build on top | Consumer of your existing tools | Agents build, host, and maintain products on the same server they run on |
| Pricing | Bundled: SuperGrok Heavy, Cursor Ultra $200/mo, Cursor Teams Premium $120/seat/mo | $299–$399/mo per server, bring your own model access |
| Availability | Beta: desktop and iOS, Android coming, enterprise waitlist | In production since early 2026 |
What Grok Bot actually is
Grok Bot launched in beta on 11 August 2026. Each Bot gets its own cloud computer, signs into the same apps, inboxes, and websites a person would use, and keeps working after you close your laptop, coming back when something needs approval. You can teach a Bot by having it watch you do a task once; it saves the steps as a routine. Bots keep context across conversations, and multiple Bots run in parallel and coordinate in group chats. xAI's launch line was that agents get you to 90% and "Grok Bot can finish the swing."
It is not sold standalone. It comes bundled inside SuperGrok Heavy, Cursor Ultra at $200 a month, and Cursor Teams Premium at $120 per seat per month, with enterprise access behind a waitlist. At launch xAI published no model-choice option and no self-hosting or data-governance details.
Credit where due: this is a real product from a frontier lab, the teach-by-watching flow is a genuinely good idea, and the launch put always-on agent machines in front of a mainstream audience.
What CRHQ is
CRHQ provisions a dedicated, managed server per customer and installs the harness on it: pre-built agents with working knowledge bases, a versioned library of skills and recipes, delegation between agents, scheduling, a browser, a file system, encrypted credentials, and a web interface that works on any device. You bring your own model access, and CRHQ routes each task across Claude, Codex, Grok, and Gemini, with open-source providers as an option when the cost-to-intelligence ratio favors them.
The server is the point. Your agents have full access to the machine they live on, so they can build, host, and maintain your own tools, apps, and products right there. generatespecs.com works this way today: the product is hosted on a CRHQ server, managed by agents living on that same server, and its agentic features invoke the harness they share it with.
The architecture difference, in plain terms
Grok Bot's unit is an agent with a computer inside xAI's walls. You interact with the agent; xAI owns the machine, the models, and the surrounding stack.
CRHQ's unit is a server with a fleet inside your walls. CRHQ manages and secures the machine the way a hosting company would; everything on it answers to you. Three things follow:
No dependency
Each task runs on whichever model earns it. When a better or cheaper model ships, from any lab or from the open-source world, your fleet switches. There is no single vendor whose roadmap decides yours.
Building is easy
Your agents can install what they need, extend their own capabilities, and host entire products on the machine where they operate, with full access.
Data governance
Your data lives on one dedicated machine. You decide what enters and leaves it. CRHQ runs the security operations; the data is governed by you.
Side-by-side
| Capability | Grok Bot | CRHQ |
|---|---|---|
| Compute model | One cloud computer per Bot, in xAI's cloud | One dedicated server per customer, whole fleet on it |
| Model selection | xAI's stack; no published choice | Claude, Codex, Grok, Gemini per task; open-source optional |
| Persistent context | Yes, across conversations | Yes: memory, project documents, files, browser login state |
| Repeatable routines | Teach by watching, saved as routines | Sessions become recipes; recipes run on schedules |
| Multi-agent coordination | Bots message each other, share threads | Delegation: an orchestrator agent runs specialists in parallel, tracked to completion |
| Agent knowledge | Learns preferences over time | Knowledge bases you can open, read, and edit: files, folders, workflows |
| Build and host your own products | No | Yes, on the same server (generatespecs.com is the live reference) |
| Server access | None published | Full access for you and your agents |
| Data location | xAI's cloud | Your dedicated server |
| Skills library | Routines you teach | ~25 versioned platform skills plus your own, shared across the fleet |
| White-label | No | Yes, partners ship it as their own product |
| Support | Enterprise waitlist | Onboarding and support from people who run fleets of their own |
Where Grok Bot wins
- You might already be paying for it. If you hold SuperGrok Heavy or Cursor Ultra, Grok Bot arrives at no extra cost. That is a distribution advantage CRHQ does not have.
- Zero decisions to start. No model access to bring, no setup conversation. Sign in and go.
- Teach-by-watching is a lovely onboarding flow for people who have never written an instruction file.
- Deep X and Grok ecosystem integration, with xAI's engineering weight behind it.
If you live inside the Grok and Cursor ecosystem, want agents from that stack, and your data leaving for xAI's cloud is acceptable, Grok Bot is a solid pick.
Where CRHQ wins
- Model freedom. Four frontier labs and the open-source world, routed per task. Grok Bot commits you to one lab's stack.
- A machine you actually control. Full access, for you and for your agents. This is what makes CRHQ buildable: agents extend their own capabilities and host real products on the server they run on.
- Data governance with a straight answer. Your data sits on one dedicated machine. Where Grok Bot's launch materials are silent on governance, CRHQ's answer fits in a sentence.
- Everything arrives assembled. Pre-built agents with working knowledge bases, a versioned skills library, delegation, schedules, and onboarding from a human. The setup work is done before you log in.
- The elevate path. Run work with an agent, have the agent capture it as a recipe, hand the recipe to a fresh agent, put it on a schedule, grow it into a team. Builders on CRHQ climb this ladder every week, and the CRHQ Community shares what works.
- White-label. Partners ship CRHQ as their own product; their clients get a server plus a product instance.
Can Grok Bot do what CRHQ does?
For personal task automation inside its ecosystem, largely yes, and comfortably. For the rest, the architecture decides. A Bot cannot give you its computer, so you cannot host your product on it, point your domain at it, choose a different lab's model for a task, or answer a client's "where exactly is our data" with a machine you control. Those are properties of owning the server, and owning the server is the part xAI keeps.
Pricing
| Grok Bot | CRHQ | |
|---|---|---|
| Entry | Bundled with Cursor Ultra $200/mo | $299/mo (Standard server) |
| Team | Cursor Teams Premium $120/seat/mo | $349/mo (Pro server) |
| Heavy | SuperGrok Heavy bundle | $399/mo (Premium server) |
| Model usage | Included, xAI's stack | Bring your own API keys, any supported provider |
| Enterprise | Waitlist | Talk to us |
If you already pay for the bundles, Grok Bot is effectively free. CRHQ costs more at entry and buys you the server, the model freedom, and everything pre-assembled on it.
Bottom line
- Pick Grok Bot if you are in the Grok or Cursor ecosystem already, want capable always-on agents with zero setup, and are comfortable with your agents and data operating inside xAI's cloud.
- Pick CRHQ if you are a builder: you want the fleet on a server you control, the freedom to run any frontier or open-source model per task, agents that can build and host your own products, and a clear answer on where your data lives.
The category now has two answers to "give every agent a machine." One of them hands you the keys.