xAI open-sources Grok Build for terminal coding

Grok Build is xAI's terminal-based AI coding agent and fullscreen terminal UI (TUI). It is built to understand your codebase, edit files, execute shell commands, search the web, and tackle long-running tasks right from your command line.

Beyond fullscreen interactive use, it also supports a headless mode for automation scripts and CI pipelines, and can be embedded directly into code editors using the Agent Client Protocol (ACP).

What it does and how it runs

The codebase gives you the full agent runtime harness. That includes context assembly, model response parsing, and tool-call dispatch for file editing, searching, and running terminal commands.

The harness is highly extensible. It supports skills, plugins, lifecycle hooks, subagents, and Model Context Protocol (MCP) servers. If you want to link private APIs or custom toolkits, you can configure them cleanly.

Best of all, Grok Build can run fully local-first. You can compile the binary yourself, point it to your own local inference backend, and handle all your preferences through a single config.toml file.

Getting started and code access

To try it immediately, xAI provides prebuilt binaries for macOS, Linux, and Windows via an installation script, which installs the command as grok.

If you prefer building from source, the project is written in Rust. You will need the toolchain pinned in rust-toolchain.toml, along with DotSlash to manage internal tools like protoc. macOS and Linux are the primary supported build environments, while Windows builds from source are currently best-effort.

The first-party code is published on GitHub under the Apache 2.0 license. One important catch: xAI notes that external pull requests are not currently accepted, as the public repository is periodically mirrored from their internal monorepo.

Why it matters

If you spend your day in terminal windows or want to inspect how frontier labs build robust agent loops, Grok Build is worth a close look. Showing the exact plumbing for context dispatch and MCP integration removes a lot of guesswork for engineers building their own agent frameworks.

That said, the no-external-contributions policy makes it clear this is a source-available reference harness rather than a community-driven project. It is great for self-hosting and studying, but do not expect your PRs to make it upstream.