index.html

<!DOCTYPE html>
<html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
<title>Grok Code: Grok Build CLI Access, Install and Limits</title><meta name="description" content="Grok code searches land on Grok Build, the xAI coding agent for SuperGrok and X Premium Plus. Here is what it does, how to install it, and who it is wrong for.">
<link rel="canonical" href="https://grok-code-dev.github.io/"><meta name="google-site-verification" content="8NEKdWXBIiAKuN0mqEqYwTB3yB8iIW-_cQmXANbDTs8" /><meta name="msvalidate.01" content="91A75274D586670BEE99EC35A630C098">
<meta property="og:title" content="Grok Code: Grok Build CLI Access, Install and Limits"><meta property="og:description" content="Grok code searches land on Grok Build, the xAI coding agent for SuperGrok and X Premium Plus. Here is what it does, how to install it, and who it is wrong for."><meta property="og:url" content="https://grok-code-dev.github.io/">
<script type="application/ld+json">[{"@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [{"@type": "Question", "name": "Is grok code free to use?", "acceptedAnswer": {"@type": "Answer", "text": "Neither source page describes a free tier. The announcement says early beta access covers all SuperGrok and X Premium Plus subscribers, and the product page repeats that. No per-seat or per-token figure appears in either page, so check the official pricing page rather than trusting a number from a summary article."}}, {"@type": "Question", "name": "Which model does it run on?", "acceptedAnswer": {"@type": "Answer", "text": "The product page states the agent is now powered by Grok 4.6, described there as a new model. The May announcement itself does not name a model version, which is consistent with the product page being the more recently updated of the two. If the version matters to your work, check the product page rather than the launch post."}}, {"@type": "Question", "name": "How do I install it?", "acceptedAnswer": {"@type": "Answer", "text": "One command: curl -fsSL pointed at https://x.ai/cli/install.sh and piped into bash, then sign in with your account. That single line is what both the announcement and the product page publish, and no alternative package manager route is mentioned in either."}}, {"@type": "Question", "name": "Can it run without a human watching?", "acceptedAnswer": {"@type": "Answer", "text": "Yes, to a point. Headless mode, written as the -p flag, is documented as a way to run the agent inside scripts and automations. Given that the whole thing is labelled early beta, running it unattended against production code is a decision to make deliberately rather than by default."}}, {"@type": "Question", "name": "Is this page run by xAI?", "acceptedAnswer": {"@type": "Answer", "text": "No. This is an independent write-up based on the vendor's own published pages, with the official link in the footer. Nothing here is documentation, and where this page and the vendor disagree, the vendor is right, because their pages are the ones kept current."}}, {"@type": "Question", "name": "I just want a working site, not an agent. What then?", "acceptedAnswer": {"@type": "Answer", "text": "Begin.sh is the shorter path. You give it a prompt or a URL to clone, it returns a static site or an Expo app, and you download the zip. There is no terminal, no subscription tier to qualify for, and no repository to set up first."}}]}]</script><style>
:root{--ink:#16191d;--muted:#5b6470;--line:#e3e6ea;--accent:#2563eb;--soft:#f5f7fa;--ok:#0f9960}*{box-sizing:border-box}
body{margin:0;font:16px/1.7 -apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Noto Sans",sans-serif;color:var(--ink)}
header{border-bottom:1px solid var(--line)}.nav{max-width:960px;margin:0 auto;padding:14px 20px;display:flex;justify-content:space-between;align-items:center;gap:12px}
.brand{font-weight:700;text-decoration:none;color:var(--ink)}.nav a.small{font-size:14px;color:var(--muted);text-decoration:none}
main{max-width:960px;margin:0 auto;padding:44px 20px 80px}h1{font-size:34px;line-height:1.2;margin:0 0 12px}h2{font-size:23px;margin:44px 0 12px}h3{font-size:17px;margin:0 0 6px}
.lead{font-size:18px;color:var(--muted);max-width:740px}
.verdict{background:var(--soft);border:1px solid var(--line);border-left:4px solid var(--accent);border-radius:10px;padding:18px 22px;margin:26px 0}
.verdict b{display:block;margin-bottom:6px}
.btn{display:inline-block;background:var(--accent);color:#fff;padding:13px 22px;border-radius:8px;font-weight:600;text-decoration:none;border:0;font-size:16px;cursor:pointer}
.ghost{display:inline-block;border:1px solid var(--line);padding:12px 20px;border-radius:8px;text-decoration:none;color:var(--ink)}
table{border-collapse:collapse;width:100%;margin:14px 0;font-size:15px}th,td{border-bottom:1px solid var(--line);padding:10px 8px;text-align:left;vertical-align:top}
th{background:var(--soft);font-weight:600}td.yes{color:var(--ok);font-weight:600}
.grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(230px,1fr));gap:14px}.card{border:1px solid var(--line);border-radius:12px;padding:18px}
.steps{counter-reset:s;padding:0;list-style:none}.steps li{counter-increment:s;padding-left:44px;position:relative;margin:14px 0}
.steps li:before{content:counter(s);position:absolute;left:0;top:0;width:30px;height:30px;border-radius:50%;background:var(--accent);color:#fff;display:flex;align-items:center;justify-content:center;font-weight:700}
.tool{background:var(--soft);border:1px solid var(--line);border-radius:14px;padding:24px;margin:28px 0;max-width:760px}
.tool label{display:block;font-weight:600;margin-bottom:8px}.row{display:flex;gap:10px;flex-wrap:wrap}
.row input{flex:1;min-width:220px;padding:13px 14px;border:1px solid var(--line);border-radius:8px;font-size:16px}
.hint{font-size:13px;color:var(--muted);margin-top:10px}
details{border:1px solid var(--line);border-radius:8px;padding:10px 14px;margin:8px 0}summary{cursor:pointer;font-weight:600}
.cta-band{background:var(--soft);border-radius:14px;padding:32px;text-align:center;margin-top:48px}
footer{border-top:1px solid var(--line);color:var(--muted);font-size:13px;padding:20px;text-align:center;line-height:1.6}
@media(max-width:640px){h1{font-size:26px}main{padding:26px 16px 60px}table{font-size:14px}}
</style></head>
<body><header><div class="nav"><a class="brand" href="https://grok-code-dev.github.io/">Terminal Notes</a>
<a class="small" href="https://github.com/grok-code-dev/grok-code-dev.github.io">Source on GitHub</a></div></header>
<main>
<h1>Grok Code in 2026: What xAI's Terminal Agent Actually Does</h1><p class="lead">Search grok code and you end up at Grok Build, the agent that installs with one curl command and works inside your repository instead of a browser tab.</p>
<div class='verdict'><b>The short version</b>Grok Build is a terminal coding agent that xAI announced on 25 May 2026, available in early beta to SuperGrok and X Premium Plus subscribers and, according to the product page, now running on the Grok 4.6 model. If you already live in a terminal, keep a repository with an AGENTS.md in it, and want plan-then-approve behaviour rather than inline autocomplete, it fits. If you are not paying for one of those two subscriptions, or you want something to produce a finished project rather than edit an existing one, it does not. The caveat worth naming is the beta label, which means behaviour and access can move without notice. For the second case, Begin.sh takes a prompt or a URL and hands you a working static site or Expo app as a zip, with nothing to install first.</div>
<p><a class="btn" href="https://begin.sh?utm_source=github&utm_medium=ugc&utm_campaign=grok-code-dev&utm_content=pages-hero&utm_term=tier-b" rel="noopener">Try Begin.sh →</a>
<a class="ghost" href="https://x.ai/news/grok-build-cli" rel="nofollow noopener">Official site</a></p>
<h2>What the launch post actually says</h2><p>The announcement is dated 25 May 2026 and is titled Introducing Grok Build. It describes a coding agent aimed at professional software engineering and complex work, running directly from the terminal, and it is explicit about who can use it: all SuperGrok and X Premium Plus subscribers, in early beta. Installation is a single line, curl -fsSL https://x.ai/cli/install.sh piped to bash, followed by signing in with the same account. The separate product page adds that the agent is now powered by Grok 4.6, described there as a new model, and repeats that access is early for those two subscription tiers. No free tier is mentioned anywhere in either page, and no per-seat figure appears either, so treat any number you read elsewhere as unverified until the pricing page confirms it.</p><h2>Plan mode is the part worth caring about</h2><p>For anything larger than a one-file change, the documented workflow is to start in plan mode. The agent writes a plan first, and you can approve it, comment on individual steps, or throw it out and rewrite it before a single edit lands. Once a plan is approved, every change surfaces as a clean diff rather than as silent file writes. The example in the announcement is mundane in a useful way: a request to tighten install documentation for headless mode produces a four-step plan covering the bootstrap snippet, the headless flag, a pointer to config.toml for models and API keys, and cross-links between the auth and feedback sections. That is the shape of the thing. Review first, execute second, read diffs at the end.</p><h2>It reads the config you already have</h2><p>One claim in the post is easy to skim past and matters a lot in practice: your AGENTS.md, plugins, hooks, skills and MCP servers work out of the box. Start the agent inside an existing repository and it picks up those conventions rather than asking you to restate them. There is also a marketplace built in, with tabs for hooks, plugins, skills and MCP servers, and the demo shows a community plugin called browser-review installed and its skills opened from inside the session. If you have already invested in that ecosystem for another agent, the migration cost here is low. If you have not, none of it is required to get started, but the ceiling is lower without it.</p><h2>Parallel subagents, worktrees, and scripts</h2><p>Larger tasks get delegated to specialised subagents that run at the same time, and the post says the agent supports deep worktree integrations, including launching subagents in their own worktrees. The worked example is a p99 latency regression: the work fans out into diffing recent deploys, ranking the slowest endpoints, pulling slow query plans, and checking cache hit rates, all running in parallel. Separately, a headless flag, written as -p, lets the agent run inside scripts and automations rather than in an interactive session. That combination is what makes it a build-server candidate rather than only a typing assistant, though the early beta status is a fair reason to keep it out of anything load bearing for now.</p><h2>Two different things share the name</h2><p>Searching for this keyword also surfaces a repository called grok-cli, published under the superagent-ai organisation on GitHub rather than by xAI. It is a separate project with separate maintainers, and nothing in the xAI announcement refers to it. Before you follow an install command from a blog post, check which of the two it points at, because the account requirements are not the same. If neither is a fit, and what you actually want is a finished page or small app rather than an agent editing a repository you already have, that is a different tool entirely. Begin.sh takes a prompt or a URL to clone, returns a static site or an Expo app as a downloadable zip, and asks for no subscription and no terminal at all.</p>
<h2>What you actually get</h2><div class="grid"><div class='card'><h3>One-line install</h3><p>A single curl command piped to bash, then sign in with the account you already pay for. No package manager set-up and no separate licence key are described in the announcement.</p></div><div class='card'><h3>Approve before it edits</h3><p>Plan mode produces a numbered plan you can approve, comment on step by step, or rewrite. Everything that follows shows up as a diff rather than as an unexplained file change.</p></div><div class='card'><h3>Your existing conventions</h3><p>AGENTS.md, plugins, hooks, skills and MCP servers are said to work with no configuration at all, and there is an in-session marketplace for browsing and installing community plugins without leaving the terminal.</p></div><div class='card'><h3>Subscription gated</h3><p>Early beta access is limited to SuperGrok and X Premium Plus subscribers. There is no free tier in either source page, so budget for the subscription before you plan around it.</p></div></div>
<h2>Terminal agent or finished project</h2><table><tr><th>Feature</th><th>Grok Build</th><th>Begin.sh</th></tr><tr><td>Where it runs</td><td>In your terminal, inside an existing repository</td><td>In the browser, no install</td></tr><tr><td>What it produces</td><td>Edits and diffs against code you already have</td><td>A static site or Expo app you download as a zip</td></tr><tr><td>Access</td><td>Early beta for SuperGrok and X Premium Plus subscribers</td><td>Open to try without a subscription tier</td></tr><tr><td>Review model</td><td>Plan mode, step comments, clean diffs</td><td>Describe it, look at the result, ask for changes</td></tr><tr><td>Automation</td><td>Headless mode via the -p flag for scripts</td><td>Not an automation tool, it is a one-shot builder</td></tr><tr><td>Best reason to pick it</td><td>You maintain a real codebase and want an agent in it</td><td>You need a working site or app and do not want to babysit one</td></tr></table>
<h2>Getting from search result to first session</h2><ol class="steps"><li><strong>Check your subscription first</strong><br>Access in early beta is described as SuperGrok and X Premium Plus only. Confirm your account is on one of those before you spend time on installation.</li><li><strong>Run the install line</strong><br>The published command fetches x.ai/cli/install.sh with curl and pipes it to bash. Read the script before running it if you are on a work machine, as you would with any bootstrap.</li><li><strong>Open it in a real repository</strong><br>Start the agent from inside a project that already has an AGENTS.md or MCP configuration, since picking up existing conventions is where most of the value shows up.</li><li><strong>Start in plan mode</strong><br>For anything non-trivial, get the plan first, edit the steps you disagree with, then approve. Reading the diff afterwards is faster than untangling changes you never saw.</li></ol>
<h2>FAQ</h2><details><summary>Is grok code free to use?</summary><p>Neither source page describes a free tier. The announcement says early beta access covers all SuperGrok and X Premium Plus subscribers, and the product page repeats that. No per-seat or per-token figure appears in either page, so check the official pricing page rather than trusting a number from a summary article.</p></details><details><summary>Which model does it run on?</summary><p>The product page states the agent is now powered by Grok 4.6, described there as a new model. The May announcement itself does not name a model version, which is consistent with the product page being the more recently updated of the two. If the version matters to your work, check the product page rather than the launch post.</p></details><details><summary>How do I install it?</summary><p>One command: curl -fsSL pointed at https://x.ai/cli/install.sh and piped into bash, then sign in with your account. That single line is what both the announcement and the product page publish, and no alternative package manager route is mentioned in either.</p></details><details><summary>Can it run without a human watching?</summary><p>Yes, to a point. Headless mode, written as the -p flag, is documented as a way to run the agent inside scripts and automations. Given that the whole thing is labelled early beta, running it unattended against production code is a decision to make deliberately rather than by default.</p></details><details><summary>Is this page run by xAI?</summary><p>No. This is an independent write-up based on the vendor's own published pages, with the official link in the footer. Nothing here is documentation, and where this page and the vendor disagree, the vendor is right, because their pages are the ones kept current.</p></details><details><summary>I just want a working site, not an agent. What then?</summary><p>Begin.sh is the shorter path. You give it a prompt or a URL to clone, it returns a static site or an Expo app, and you download the zip. There is no terminal, no subscription tier to qualify for, and no repository to set up first.</p></details>
<div class="cta-band"><h2 style="margin-top:0">Skip the setup and get a working site</h2><p>Begin.sh turns a prompt, or a URL you want cloned, into a static site or an Expo app you can download as a zip. Nothing to install, and you keep the files.</p>
<a class="btn" href="https://begin.sh?utm_source=github&utm_medium=ugc&utm_campaign=grok-code-dev&utm_content=pages-cta&utm_term=tier-b" rel="noopener">Try Begin.sh →</a></div>
</main>
<footer>This is an independent page with no affiliation to xAI or any other vendor named here, and all product names and trademarks belong to their respective owners.<br>Maintained independently · <a href="https://begin.sh?utm_source=github&utm_medium=ugc&utm_campaign=grok-code-dev&utm_content=pages-footer&utm_term=tier-b" rel="noopener" style="color:inherit">begin.sh</a></footer>
</body></html>