SpaceXAI Launches Grok Bot Beta: Persistent AI Agents That Sign Into Your Apps

SpaceXAI shipped the public beta of Grok Bot on August 11, 2026, giving a first look at what the $60 billion Cursor acquisition was actually supposed to build. The product started as an internal xAI prototype. According to the company, it has already changed how the xAI team works day to day.

The pitch is straightforward: instead of a chatbot that answers questions, Grok Bot is an agent that does work. Each bot gets its own computer, signs into the apps and services the user already uses, and operates across inboxes, calendars, and other tools on its own. It only surfaces when it needs a human decision. Multiple bots can coordinate with each other, passing context and handing off tasks, and the system learns preferences over time.

Who Gets Access

The beta is limited. You need one of three subscriptions to get in:

SuperGrok Heavy ($300 per month)
Cursor Ultra ($200 per month)
Cursor Teams Premium ($120 per seat per month)

An enterprise waitlist is also open. Android support is coming; Mac, iOS, Windows, and Linux are live now.

The company is keeping access tight during this rollout because it overlaps with the organizational work of integrating Anysphere into SpaceX. Elon Musk followed up after the launch announcement to confirm a wider rollout is coming, tied to Grok 4.6 going broader.

This Is the First Real SpaceX-Cursor Joint Product

SpaceX announced the Cursor acquisition in June 2026 at a $60 billion all-stock valuation. That made it the largest startup acquisition ever, and it prompted a predictable wave of coverage about what the combined company would actually produce. Grok Bot is the answer.

The connection is direct. xAI’s model development, SpaceX’s compute, and Cursor’s distribution and tooling are all in play. Rather than bolting an AI assistant onto the Cursor interface, the team built an autonomous agent platform that uses apps the way a human would — through the UI, not the API. That matters for the long tail of tools that don’t have an API, or that have APIs too locked down for third-party agents to reach.

xAI’s internal case studies include overnight sales research, expense receipts pulled from email, and bug fixes — straightforward workflows that are easy to scope and verify, which makes them reasonable places to start.

The Security Question

SpaceXAI has had credentialing problems before. Grok Build, the earlier coding agent, was found to be uploading entire Git repositories to xAI’s servers including any committed secrets. The company patched it quickly, but the incident raised questions about how autonomous agents handle credentials when working across apps.

Grok Bot’s model — agents that sign into apps as you — puts that question front and center. SpaceXAI hasn’t published detailed documentation on how it sandboxes credentials or isolates sessions between bots. That’s one of the things to watch as access widens.

What’s Coming Next

Musk confirmed a broader rollout, timed with Grok 4.6. The model launched August 7 and uses the same 1.5-trillion-parameter V9 architecture as Grok 4.5, with improved supervised fine-tuning and reinforcement learning. Grok 4.7, a larger 2.1-trillion-parameter model, is expected in late August or early September.

For now, Grok Bot is the most direct signal yet of where the SpaceXAI-Cursor roadmap is heading: away from the IDE and toward the desktop, inbox, and calendar.

Sources: