Grok Build
FreemiumSpaceXSI's coding agent harness for the terminal.
Published 27 September 2026Last updated 3 November 2026
Scores
Popularity2/5
Emerging
A three-month-old beta product at time of evaluation with a subscription-gated, non-self-hostable adoption path, giving it a narrower reach than the established players in this subcategory.
Learning Curve2/5
Easy
Automatic recognition of existing AGENTS.md/CLAUDE.md conventions and a familiar plan/review/approve loop make onboarding straightforward for developers coming from other CLI agents.
Flexibility3/5
Configurable
Headless mode, full ACP support, and worktree-isolated parallel sub-agents provide real configurability, though the underlying model and billing are fixed to xAI's own subscriptions with no bring-your-own-key path.
Performance4/5
Fast
Backed by the frontier Grok 4.6 model with a 500K-token context window and a distinctive worktree-isolation mechanic for parallel work, though vendor-reported benchmarks place it behind Claude Code/Opus 4.7 and GPT-5.5 as of evaluation.
Portability1/5
Locked in
Cloud-only, closed-source, and bundled exclusively into SuperGrok Heavy/SuperGrok/X Premium+ subscriptions, with no self-hosted or standalone API-key path. One of the least portable options in the subcategory.
About Grok Build
Grok Build (command grok) is SpaceXSI's terminal coding agent, in the same family as Claude Code and Codex CLI and separate from the general-purpose Grok Bot. Written in Rust, it runs as a full-screen, mouse-interactive terminal interface that reads the codebase, edits files, runs shell commands, searches the web, and manages long-running background tasks, with a plan mode that proposes steps for approval before anything changes. A headless mode handles scripts and CI, and the Agent Client Protocol lets editors such as Zed, Neovim, and Emacs, or a WebSocket client, drive it as a server.
Its signature is parallel subagents: the main agent hands research, implementation, testing, and review to child sessions with their own context windows, and any of them can work in an isolated git worktree that merges back when done, while an agent dashboard tracks several top-level sessions at once. It picks up existing project conventions (AGENTS.md, CLAUDE.md, hooks, skills, plugins, and MCP servers), and an optional sandbox confines commands with Landlock on Linux or Seatbelt on macOS.
Hosted use signs in with an account on the X Subscription (limited on the free plan, full from Plus up) and runs Grok 4.x models or the coding-focused Grok Build 0.1; an API key from SpaceXSI's console bills the same models per token instead. Custom endpoints speaking the OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages protocol add other providers or local inference.
SpaceXSI open-sourced Grok Build under Apache 2.0 in July 2026, days after researchers showed that earlier versions had uploaded entire Git repositories to its cloud without clear consent, and said it would delete the retained data. The public repository is a periodic sync of SpaceXSI's internal monorepo and doesn't accept outside contributions.
Key Features
- Full-screen, mouse-interactive terminal interface with plan mode and inline diffs
- Parallel subagents with their own context windows, each optionally in an isolated git worktree
- An agent dashboard for watching and steering several sessions at once
- Headless mode for scripts and CI, plus the Agent Client Protocol for Zed, Neovim, Emacs, and WebSocket clients
- Reads AGENTS.md, CLAUDE.md, hooks, skills, plugins, and MCP servers without extra setup
- Grok 4.x and Grok Build 0.1 by default, plus OpenAI- or Anthropic-compatible endpoints and local inference
- Optional sandbox using Landlock on Linux and Seatbelt on macOS
- Open source under Apache 2.0, written in Rust
Pros
- Parallel subagents and worktree isolation are built in rather than bolted on
- Native CLAUDE.md and AGENTS.md support lowers the switching cost from Claude Code
- Open source under Apache 2.0, and able to run fully local against your own inference
- Grok Build 0.1 is inexpensive per token for a coding-tuned model
- Included with X Subscription rather than sold as a separate product
Cons
- Earlier versions uploaded whole repositories to SpaceXSI's cloud without clear consent, a trust issue for sensitive code
- Younger than Claude Code or Codex, with a smaller community and plugin ecosystem
- Outside contributions aren't accepted; the public repository is a periodic sync of SpaceXSI's monorepo
- No official way to steer a running session from a phone or browser; only community tools fill that gap
- Easily confused with SpaceXSI's separate Grok Bot product
Grok Build Pricing
Freemium- Apache 2.0 source on GitHub
- Bring your own local inference or API key
- Limited use on the free plan; full use from Plus ($8) up
- Shares the plan's usage pool with Grok and Cursor
- Grok Build 0.1: $1/M input, $2/M output under 200K tokens, double above
- Grok 4.7: $2/M input, $6/M output under 200K tokens, double above
- Billed through the SpaceXSI console with an API key
Tech Stacks with Grok Build
Grok Build Dev Stack
Developer
DeveloperGrok Build as the coding agent in the terminal, running Grok or models such as Claude and GPT, with parallel subagents, GitHub or GitLab for version control, and optional local models through Ollama.
Development
LLM
Orca Dev Stack
DeveloperOrca runs several coding agents side by side, each in its own git worktree, with your pick of agents and models underneath and GitHub or GitLab for version control.
Herdr Dev Stack
DeveloperHerdr keeps several coding agents running in persistent terminal sessions on your laptop or a server, with your pick of agents and models and GitHub or GitLab for version control.
Tools Related to Grok Build
Works well with Grok Build(2)
Paradev runs Grok Build sessions in parallel git worktrees and tracks their quota and token use on its Usage page.
T3 Code adds a GUI, worktree isolation, and mobile/desktop clients on top of Grok Build's CLI for developers running it alongside other coding agents.
Integrates with Grok Build(1)
Grok Build is SpaceXSI's own coding agent for its models: a sign-in on an X Subscription plan runs Grok 4.x or the coding-tuned Grok Build 0.1, and a SpaceXSI API key bills the same models per token.
Alternatives to Grok Build(12)
OpenCode is the largest model-agnostic open-source agent across terminal, IDE, and desktop; Grok Build is SpaceXSI's terminal agent running parallel worktree sub-agents on Grok or other models. OpenCode has the bigger ecosystem, Grok Build built-in parallelism.
Grok Build focuses on parallel sub-agents in separate git worktrees on Grok models; oh-my-pi packs 31 tools, LSP/DAP, and browser automation across 60+ providers. Grok Build for parallelism, oh-my-pi for tool breadth.
GitHub Copilot lives in github.com and every major IDE with completions, chat, and agent mode; Grok Build is a terminal agent running parallel sub-agents in git worktrees on Grok. Copilot wins on reach, Grok Build on parallel CLI runs.
Cline runs in VS Code and the CLI with step-by-step Plan/Act approval; Grok Build is a full-screen terminal agent that fans work out to parallel subagents, optionally in separate git worktrees. Cline suits supervised editing, Grok Build parallel work on Grok models.
Grok Build reads Claude Code's CLAUDE.md, hooks, and skills natively, so a project moves over without rework. Grok Build is open source and runs Grok or any compatible model; Claude Code adds hosted web sessions and phone Remote Control.
Grok Build is a terminal agent running parallel sub-agents in git worktrees on Grok models; ZCode is Z.ai's desktop app tuned for GLM, with Goal Mode and chat-app remote control. Grok Build suits CLI-heavy parallel work, ZCode a desktop GUI.
