SpaceXAI, formerly xAI, introduced Grok Bot as an early beta on August 11, 2026. The company describes these bots as "AI teammates that do real work for you," capable of signing into and using applications and websites in a manner similar to a human user. Each Grok Bot operates within its own persistent cloud-based computer environment, allowing tasks to continue even when a user's local device is closed. This distinguishes Grok Bot from traditional AI assistants that typically require continuous user interaction.
Thesystem allows users to create multiple bots, assign them specific roles, and have them work concurrently. SpaceXAI has provided examples of internal use cases, including bots for sales outreach, marketing campaigns, office operations, and bug fixes. A sales bot, for instance, can research accounts, score contacts, draft outreach emails and LinkedIn messages in the user's voice, and then compile the results for human review.
A key feature of Grok Bot is its ability to learn workflows through demonstration. Users can perform a task while a bot observes, and the bot can then save that process as a routine to execute later. These learned routines can also be updated with corrections as the user provides further guidance. Bots can also collaborate, with multiple agents placed in the same thread to pass work and share context. SpaceXAI demonstrations have shown specialized bots, such as Research, Communications, Chief of Staff, and Travel Bots, coordinating tasks and messaging each other independently within threads or group conversations.
Grok Bot is accessible on macOS, Windows, Linux, and iOS, with an Android version planned for future release. Access to the beta is currently limited to subscribers of SuperGrok Heavy, Cursor Ultra, and Cursor Premium Teams. This initial limited rollout coincides with the ongoing merger between SpaceXAI and Cursor, which SpaceX acquired for $60 billion in June. Pricing for Grok Bot starts at $120 per seat per month for teams and $200 per month for individuals.
The introduction of Grok Bot positions SpaceXAI in an increasingly competitive market for AI agents. Other companies, including Anthropic and OpenAI, have also developed AI agents capable of interacting with computer interfaces and external applications. Anthropic introduced computer use for Claude in 2024 and expanded its capabilities with Claude Code and Claude Cowork agents in 2025 and 2026, respectively. OpenAI also enabled its Codex harness to control other computer applications in April 2026 and launched agentic Workspace Agents.
A notable aspect of Grok Bot's architecture is that all bots associated with a single user account share the same persistent cloud computer. This shared environment means that browser sessions, files, and application logins are accessible to all bots under that user's account. While each bot receives its own screen on this shared machine, SpaceXAI clarifies that these are separate work surfaces, not separate security boundaries. This design choice allows for seamless handoffs between bots without redundant setup, but it also implies that any login or file placed on the computer becomes available to all bots.
SpaceXAI states that Grok Bot uses Cursor authentication, single sign-on (SSO), and privacy mode. The cloud computer employed by Grok Bot is encrypted both in transit and at rest, and users have an option to opt out of training data collection. For sensitive actions, the system can route them through an "Auto Review" process before execution. Enterprise administrators are offered controls for data-loss prevention, certificates, proxies, and network configurations.
The launch of Grok Bot comes amid previous scrutiny regarding Grok's safety and privacy. In December 2025, reports indicated that Grok had revealed personal addresses and provided instructions for stalking. In February 2026, federal agencies reportedly raised concerns about Grok's reliability and susceptibility to manipulation before its approval for use in classified settings by the Trump administration. Additionally, a digital safety nonprofit, Common Sense Media, issued a report in January 2026, stating that Grok's safeguards were inadequate for teen users and that its business model encouraged misuse, particularly concerning explicit content and biased responses. In February 2026, the European Union's data privacy regulator initiated an investigation into X following reports that Grok's AI chatbot generated nonconsensual deepfake images. Earlier, in August 2025, an estimated 370,000 private chats from Grok were accidentally published, some containing instructions for illegal activities.
SpaceXAI's official announcement emphasizes the convenience and efficiency of Grok Bot, highlighting its ability to handle repetitive administrative tasks across various departments. The company aims for the bots to reduce manual work and streamline complex workflows, allowing human employees to focus on tasks requiring judgment.