xai-org/grok-build

Grok Build is SpaceXAI's terminal coding agent, written in Rust and shipped as a full-screen TUI invoked as grok. It edits files, runs shell commands, searches the web, and manages long-running tasks — all from the terminal. Three modes are supported out of the box: interactive TUI, headless for scripting and CI, and embedded in editors via the Agent Client Protocol (ACP).

The architecture is cleanly separated across crates: the TUI layer handles rendering and scrollback, a distinct agent runtime (xai-grok-shell) manages leader, stdio, and headless entry points, and tool implementations live in their own crate (xai-grok-tools). That separation matters because headless mode and ACP integration mean grok is designed to run in CI pipelines and inside editors, not just interactively at a developer's terminal. The README lists MCP server support, plugins, hooks, and sandboxing as documented features — the scope is closer to an agent platform than a chat wrapper.

The build story is honest about its rough edges. Windows builds are described as best-effort and untested from this tree. The root Cargo.toml is generated from the internal monorepo and marked read-only, which signals that this public repo is a periodic sync rather than a primary development surface. A SOURCE_REV file tracks the upstream commit SHA, making the gap between internal and public state at least visible, if not narrow.

The licensing note is worth reading carefully. First-party code is Apache 2.0, but the third-party notices explicitly call out ported tool implementations from openai/codex and sst/opencode. Those ports are acknowledged under the Apache §4(b) change notice requirement — the right thing to do — but it also means the tool layer is partly derivative of other open-source projects rather than built entirely from scratch.

External contributions are not accepted, full stop. That makes this a source-available release for auditability and trust rather than a community project. You can read the code, build it yourself, and verify what the agent is doing on your machine — you cannot send a pull request.

For anyone evaluating terminal coding agents, the Rust implementation and clean crate boundaries are genuine differentiators. The agent runtime is not a Python script wrapping API calls. Whether SpaceXAI's model quality justifies switching from an already-working setup is a separate question the README cannot answer.