xAI published the full source code of Grok Build, its terminal-based coding agent, on GitHub under an Apache 2.0 license: roughly 844,000 lines of Rust with only about 3 percent vendored code, meaning almost the entire harness is now inspectable. The release covers four pieces that most coding-agent products keep closed: the agent loop that assembles context, parses model responses, and dispatches tool calls; the tools layer that handles reading, editing, and searching code and running shell commands; the terminal UI, including rendering, input handling, plan review, and the inline diff viewer; and the extension system covering skills, plugins, hooks, MCP servers, and subagents. For developers building or evaluating coding agents, this is unusually useful because most competing tools (Claude Code, Codex CLI, Cursor, and others) expose a polished product surface but not the underlying orchestration logic, so understanding how context gets assembled, how tool-call loops are structured, or how a diff viewer should handle partial edits has mostly required reverse-engineering from behavior. Grok Build being open means that logic is now a readable reference implementation, and because it is written to run local-first, pointed at your own inference endpoint via a config file rather than xAI's API, it also becomes a usable starting point for teams who want a terminal coding agent that does not depend on any particular model vendor. Expect it to get forked and adapted quickly, both by teams building custom internal coding agents and by other open harnesses borrowing specific pieces like the diff viewer or the MCP/subagent extension system.