Grok Build: Elon Musk’s xAI Open‑Source Full‑Screen TUI AI Agent Hits 7K Stars in Two Hours

Overview

Grok Build is an open‑source, full‑screen terminal UI (TUI) AI programming agent written in Rust. It runs without an IDE and is released under the Apache 2.0 license.

Architecture

UI layer (TUI) → Agent runtime layer → Tool execution layer

UI layer (xai-grok-pager)

Implemented with ratatui, provides a scroll buffer, input box, modal windows, syntax highlighting, inline diff rendering, mouse interaction, and keyboard shortcuts.

Agent runtime layer (xai-grok-shell)

Manages conversation state, assembles system prompts (AGENTS.md + Skills + config), calls the model, schedules tool execution, and persists sessions. Supports interactive TUI mode, stdio pipe mode, and Leader process mode.

Tool execution layer (xai-grok-tools)

Implements concrete tool logic such as file read/write, code search, shell command execution, web search, MCP protocol bridging, and memory search.

Core Features

Feature 1: Full‑screen TUI AI terminal

Scroll buffer : complete dialogue history with fold/unfold, selection, and copy.

Inline diff : line‑by‑line display of file edits.

Syntax highlighting : automatic language detection.

Modal windows : settings, search, session management.

Vim mode : navigation with j / k / h / l and copy shortcuts y / Y.

Feature 2: Rich toolset

read_file / search_replace

: line‑accurate file read/write and edit. grep: regex codebase search powered by ripgrep. list_dir: directory listing. run_terminal_command: execute shell commands. web_search / web_fetch: web search and URL fetch. todo_write: create and manage todo lists. spawn_subagent: generate parallel sub‑agents. memory_search: cross‑session memory search.

Feature 3: Claude Code ecosystem compatibility

Automatically loads CLAUDE.md, Claude.md, and CLAUDE.local.md configuration files.

Supports the .claude/rules/ directory.

Compatible with Claude Code market, plugins, skills, and MCP server.

Zero‑configuration migration.

Feature 4: Multi‑mode operation

Interactive TUI : full‑screen UI for daily development.

Headless mode ( -p): suitable for scripts and CI/CD pipelines.

ACP protocol : Agent Client Protocol embeddable in editors.

Leader process mode : agent runs in a separate process, TUI connects via WebSocket.

Feature 5: Sub‑agent system

Generates independent sub‑agents that run parallel tasks, each with its own session context and tool set.

Practical Scenarios

Scenario 1: Codebase understanding and refactoring

Run grok in the project directory.

Enter @src/ 带我了解这个项目的架构 to request an architecture analysis.

Grok reads all source files, analyses dependencies, and outputs an architecture description.

Enter 将 utils 模块拆分为独立的文件 to request refactoring.

Grok performs the split and shows inline diffs.

Scenario 2: CI/CD automated code review

grok -p "Review changes for bugs and security issues" \
  --output-format json --yolo | jq -r '.text'

Scenario 3: Parallel bug investigation

In the TUI, request creation of a sub‑agent.

Each sub‑agent investigates a different file.

Results are aggregated and displayed.

Installation

Environment requirements

macOS / Linux / Windows (Git Bash or WSL).

No Rust installation needed; pre‑compiled binaries are provided.

Installation commands

# macOS / Linux / Git Bash
curl -fsSL https://x.ai/cli/install.sh | bash

# Windows PowerShell
irm https://x.ai/cli/install.ps1 | iex

# Verify installation
grok --version

# Update to latest version
grok update

Build from source

# Requires Rust toolchain + protoc
cargo run -p xai-grok-pager-bin      # Build + launch TUI
cargo build -p xai-grok-pager-bin --release  # Release build

Run

export XAI_API_KEY="xai-xxx"
grok

Request Flow

User input prompt
   ↓
Agent builds system prompt (AGENTS.md + Skills + config)
   ↓
Model call (supports grok‑4.5 and custom models)
   ↓
Model returns tool‑call request
   ↓
ToolBridge schedules tool execution
   ├── Local tools (file edit, command exec)
   ├── MCP tools (external server)
   └── Hosted tools (server‑side search)
   ↓
Tool results returned to model
   ↓
Model generates final reply
   ↓
Result rendered in TUI scroll buffer

Project Structure

grok-build
├── Binary entry (xai-grok-pager-bin)
├── TUI layer (xai-grok-pager)
├── Agent runtime (xai-grok-shell)
├── Tool library (xai-grok-tools)
├── Agent builder (xai-grok-agent)
├── Codebase graph (xai-codebase-graph)
├── File system (xai-fast-worktree)
├── Common utilities (crates/common/)
├── Third‑party deps (third_party/)
└── Production config (prod/)
    └── mc/cli-chat-proxy-types

References

Project repository: https://github.com/xai-org/grok-build
Official documentation: https://docs.x.ai/build/overview