How Cursor's Browser Experience Shaped Grok Bot

Miller describes a fast-moving few months and says SpaceX AI first gave Cursor access to its Colossus compute system while the companies explored an acquisition. Cursor had developed models including Tab autocomplete and Composer, with the goal of building a strong model for coding. After training Composer 2.5 on Colossus, the teams began working together on Grok 4.5 and Grok 4.6.

Miller says the companies are still working out how to combine their products and brands. SpaceX AI brings voice, image, and video APIs, while Cursor focuses on coding tools, its desktop app, and cloud agents. He says the teams have had distinct areas of focus and are figuring out how existing Cursor and xAI customer plans might fit together.

Miller says model training has not yet reached a point where adding compute and training data stops improving results. He points to Cursor’s software development data, SpaceX engineering work, and X’s real-time information as resources that could benefit their models. Colossus gave Cursor capacity to pursue general-purpose Grok models, beyond its more narrowly focused Composer coding models.

Miller credits years of building Cursor and its tools, including a built-in browser launched in the desktop app during the prior year, with helping the team build Grokbot quickly. He says Grokbot reached internal beta in about a month and runs in an isolated virtual machine environment. The team treated browser interaction as essential so bots could work with websites even when an API or MCP server was unavailable.