SpaceXAI opens Grok Build after privacy uproar

SpaceXAI has open-sourced Grok Build, its terminal-based AI coding agent, after a privacy backlash over the tool’s default behavior in beta. The company published the source code on GitHub under the Apache 2.0 license on Jul. 16, 2026, giving developers a way to inspect, modify, and run the software locally.

According to SpaceXAI, the release exposes how Grok Build works end to end, including how it assembles context, dispatches tool calls, renders its terminal interface, and loads extensions such as plugins, hooks, Model Context Protocol (MCP) servers, skills, and subagents. The company said developers can also compile the tool themselves, connect it to their own local inference setup, and manage it through a local configuration file.

The move follows sharp criticism of Grok Build, which launched in May 2026 as a terminal user interface and agent runtime powered by the Grok 4.5 model. Security researchers and users found that the initial beta version defaulted to uploading entire directories to xAI’s Google Cloud servers, including sensitive files such as SSH keys, password databases, personal photos, and documents.

Recommended reading

FBI eyes Nvidia and Google TPU AI systems

In response, Elon Musk said all previously uploaded user data would be fully deleted. SpaceXAI disabled the automatic upload feature on July 12 and reset user usage quotas after the change.

The open-source release covers the coding agent’s runtime and interface, not the Grok 4.5 model itself. That means developers can audit the software around the model, but cloud-based access is still required unless they connect Grok Build to a compatible local inference setup.

That distinction matters for companies considering production use. The Apache 2.0 license allows commercial use, and the source release should make internal security reviews and compliance checks easier. But because the underlying model remains proprietary, the wider Grok platform is still only partly open.