Community-Built grok-cli Hits 3,000 Stars as Open-Source Rival to xAI's Agent
A community-built terminal coding agent wired to xAI's Grok API, with sub-agents, sandboxed verification, Telegram remote control, and X search built in.
Takeaways
- Open-source TypeScript coding agent for xAI's Grok API, 3k+ GitHub stars, MIT licensed.
- Built-in X search, web search, sub-agents, and media generation tools ship by default.
- Verify mode builds, boots, and smoke-tests your app in a sandbox with screenshots.
- Telegram remote control lets you drive the agent from your phone via voice or text.
- Optional Shuru microVM sandbox isolates shell commands from your host filesystem.
- Community-built, not affiliated with xAI's official Grok Build CLI product.
A community-run TypeScript project called grok-cli has quietly grown into one of the more feature-dense open-source coding agents on GitHub, crossing 3,000 stars and shipping a steady stream of releases. It targets xAI's Grok API and packages capabilities you'd normally cobble together from four or five separate tools: real-time search, parallel sub-agents, sandboxed verification runs, desktop automation, and remote control from a phone.
One thing to flag up front: this is a community effort that connects to xAI's Grok API, not xAI's own Grok Build CLI. The README is explicit that it isn't affiliated with or endorsed by xAI. If you're comparing options, the official product is a paid subscription tool, while grok-cli is MIT-licensed and free to fork.
Inside the toolbox
The pitch is a full-fat terminal agent rather than a chat wrapper. It bundles real-time X search, web search, the full Grok model lineup, sub-agents on by default, remote control via Telegram, and a terminal UI built with Bun and OpenTUI. The headline capabilities:
- Sub-agents by default: foreground
taskdelegation for work like exploration or general jobs, plus a backgrounddelegatefor read-only deep dives you can run in parallel. - Verify mode:
/verifyor--verifyinspects the project, figures out how to build and run it, spins up a sandbox, and produces a report with screenshots and video evidence.