SpaceXAI Releases Grok Bot, a ‘24x7’ AI Teammate

SpaceXAI has launched Grok Bot in early beta. It is an AI agent designed to operate inside users’ apps and websites and complete multi-step tasks without requiring users to supervise every action. 

The bot can run on persistent cloud computers, retain context between tasks, and return to users when an approval is needed.

Each bot runs on a persistent cloud virtual machine with its own browser, filesystem, and terminal. It can use connected tools and computer-use capabilities to operate software even when a service does not provide a dedicated API. This allows it to perform tasks directly in applications rather than simply producing instructions or drafts.

Users can create multiple bots for different jobs, such as managing an inbox, researching sales prospects, handling expenses, or reproducing software bugs. The bot can work in parallel and pass context between one another, allowing one agent to hand off part of a task to another.

Grok Bot can also learn workflows from demonstrations. 

A user can make a bot follow a multi-step process once, after which it can save the workflow as a routine and repeat it on demand or on a schedule. Bots retain their memory, files, browser sessions, and preferences, allowing work to continue across separate tasks.

SpaceXAI said Bots are designed to complete work end-to-end and only return when human input is required. Its examples include researching prospective customers, scoring accounts, drafting outreach messages, and leaving the final communications for the user to approve.

Grok Bot is available today in early beta to SuperGrok Heavy, Cursor Ultra, and Cursor Premium Teams subscribers on desktop and iOS. Enterprise customers can currently join a waitlist for access.

Grok Bot’s integration with Cursor is a result of SpaceX agreeing in June to acquire AI coding startup in an all-stock transaction valued at about $60 billion. Before the acquisition, Cursor and SpaceXAI jointly trained Grok 4.5. 

Cursor said the model was trained using trillions of tokens from Cursor data capturing developer-agent interactions and was designed for long-running work beyond software engineering.