Grok Build Open Source
Grok Build is an open-source agent harness and terminal interface built in Rust (Apache-2.0) that coordinates repository context, tools, model reasoning, edits, review, and approval. Its public repository allows developers to inspect, build from source, and extend the coding-agent workflow.
Key facts
What is Grok Build Open Source?
An inspectable, buildable, and extensible open-source agent harness that gives developers control over their AI-assisted coding workflows.
Who is Grok Build Open Source best for?
Developers, coding teams, AI engineers
What are its main limitations?
- Model inference is hosted (not open-source); weights and provider costs are separate
- Credit purchases currently in Stripe test mode – no real payments processed
- Contribution policy is closed (periodic sync from internal monorepo)
- Advanced models are locked behind credits; basic model is limited in capability
Limitations are based on product materials or editorial synthesis; confirm on the official site.
Key features
- ✓Public Rust repository (Apache-2.0 license) for local inspection and compilation
- ✓AGENTS.md repository context for project-specific instructions
- ✓Skills, hooks, MCP servers, and plugins for extensible workflows
- ✓Subagents and Git worktrees for parallel independent execution
- ✓Multiple interfaces: TUI, headless runner, and ACP client
- ✓Hosted model access via OpenRouter proxy (server-side credential, no key in browser)
- ✓Credit-based usage with 3 free questions on Grok Build 0.1 model
Use cases
- →Code generation and debugging with repository-wide context
- →Repository planning and migration (e.g., from a closed agent to Grok Build)
- →Designing and integrating MCP servers for custom tooling
- →Architecture review and reasoning with advanced models
- →Automated code review and patch generation
- →Learning and evaluating open-source agent internals
Pricing
Free
$0
- · 3 successful questions on Grok Build 0.1
- · Streaming Markdown answers
- · No card required
Builder
$10 one-time
- · 10,000 credits
- · All models including advanced
- · Suitable for occasional debugging and implementation
Pro
$25 one-time
- · 25,000 credits
- · All models
- · More room for long context and advanced Grok models
Studio
$50 one-time
- · 50,000 credits
- · All models
- · Intensive evaluation and team demos
Free plan with 3 questions; one-time credit packs from $10 (10,000 credits) to $50 (50,000 credits). No monthly subscription.
Pros
- +Open-source (Rust, Apache-2.0) – code can be inspected, built, and forked
- +Extensible with skills, hooks, MCP servers, and plugins
- +Multiple run modes (TUI, headless, ACP) for different workflows
- +Free tier available (3 questions) to test the hosted model
- +Credit system transparently priced at 1,000 credits per dollar
Cons
- −Model inference is hosted (not open-source); weights and provider costs are separate
- −Credit purchases currently in Stripe test mode – no real payments processed
- −Contribution policy is closed (periodic sync from internal monorepo)
- −Advanced models are locked behind credits; basic model is limited in capability
Positioning
- Core value: An inspectable, buildable, and extensible open-source agent harness that gives developers control over their AI-assisted coding workflows.
- Ideal for: Developers, coding teams, AI engineers
- Product type: Open-source CLI/TUI, web playground, headless runner, ACP client
Frequently asked questions
What do the first three free questions include?
They use Grok Build 0.1, the default basic model. A free question is counted only after a response is returned. Advanced models require a positive credit balance.
How are credits calculated?
One credit represents $0.001 of retail usage. The retail token rate is two times the current OpenRouter provider rate. Input and output are priced separately.
Why are advanced models locked?
Advanced models have different inference costs and are reserved for credit users. Grok Build 0.1 remains the low-cost default for testing.
Do unused credits expire or roll over?
Credits are stored on your account and do not expire during this test release. Test payments have no cash value.
Is a real payment processed on this page?
No real charge is made while the checkout is in Stripe test mode. Use the test card 4242 4242 4242 4242. Do not enter a real card.
When do purchased credits appear?
Stripe returns you to the page after payment, but the return page does not grant credits. Credits are granted via a signed Stripe webhook.
You may also like
Curated suggestions based on similar categories.
Orbit for Mac
Orbit is a native macOS desktop app that integrates an AI agent, code editor, browser, canvas, and terminal into a single environment. It supports 200+ models across 20+ providers and includes a Skills Marketplace with 20K+ community-built skills, all while keeping code local and private.
Nixmac
nixmac is an open-source macOS application that transforms your Mac into a reproducible, version-controlled system using Nix. It allows you to describe desired configurations in plain English, which it then converts into Nix code, builds, and applies safely, with real-time feedback and automatic rollback on failure.
CodeMote
CodeMote by Caste is an iOS and iPadOS app that lets you run terminal, AI coding agents, and git from your iPhone or iPad, remotely driving VS Code, Cursor, and Antigravity on your own machine without cloud copies.
Termi Protocol
The world's first 3D AI agent workspace. Termi Protocol lets developers watch their AI coding agents work live in a 3D room, with full visibility, checkpoints, task boards, and memory. It runs locally with your own API keys and works with Claude Code, Codex, Gemini CLI, and other CLI agents.
DevSnoop
DevSnoop is a local Chrome extension and native host that gives coding agents (like Claude Code, Codex, Cursor) direct browser inspection and control. It provides compact page summaries, click/fill actions, console and network logs, DOM diffs, legibility audits, and screenshots through a single HTTP API, enabling agents to verify UI changes before editing code.
What will you today?
Bolt is a vibe coding tool by StackBlitz that lets you create stunning apps and websites by chatting with AI. It integrates with design systems, provides hosting, databases, and authentication out of the box, and automatically routes to the best AI model for each task.