A developer named Aarav just launched Sprocket, an AI agent that designs hardware and writes software from the same interface, and it’s climbing fast on Hacker News with 170 points. According to the Show HN post, Sprocket is billed as a lightweight agent aimed at one big goal: becoming the best platform for building both hardware and software in one place. What stands out here is the scope. Most coding agents stop at code. This one wants to design the physical thing too.
What Sprocket Actually Does
The Hacker News listing breaks the capabilities into a few concrete areas:
- Hardware design. Sprocket generates schematics rendered in React, builds your Bill of Materials (BOM), and writes detailed assembly instructions. So it’s not just a diagram. It’s the parts list and the build guide too.
- Software development. It writes code alongside the hardware work, keeping both sides of a project in one agent instead of two tools.
- Web-sourced context. The creator says Sprocket pulls “best-in-class context from the web for everything it does,” which is the claimed basis for its reliability.
- Autonomous purchasing. This is the headline feature. Sprocket can buy things on its own when you tell it to, from hardware parts to SaaS subscriptions, across any website. Tell it what you need, and it handles the checkout.
That purchasing ability is the part worth watching. An agent that reads a BOM and then actually orders the components closes the loop between design and build. It’s also the part that should make you pause, which I’ll get to.
How to Run It
Sprocket is flexible on setup, and the docs say your choice doesn’t change performance:
- npm install:
npm i -g @spikonado/sprocket, then runsprocket. - Desktop app: Download the installer for your OS from the latest GitHub Release. Builds ship as
.AppImage,.dmg, or.exe. - Browser mode: Add the
--webflag to open Sprocket in a tab instead of the desktop app.
You can point it at a workspace directory (sprocket ../my-robot), and it remembers attached workspaces and local sessions between launches. Local state lives in $HOME/.sprocket. The desktop app uses more RAM than the browser version, per the notes.
The Technical Setup
Sprocket runs a local server (default port 17731) with the client talking to it. For developers who want to build from source, the stack is specific: Bun 1.3.9 or newer, Node.js 24.14 or newer, and a current stable Rust toolchain. It uses Convex for the agent runtime, and you configure API keys per model provider you want to enable. Development runs Vite on one port and a Rust API on another.
What to Watch Out For
The listing is honest about a few rough edges. Unsigned macOS and Windows desktop builds may trigger a Gatekeeper or SmartScreen warning the first time you open them, so you’ll need to override it manually. The CLI ships separately on npm from the desktop installers on GitHub, which means two update paths to track.
And then there’s the autonomous buying. Handing an AI agent the ability to purchase from any website is powerful and genuinely risky. There’s no mention in the post of spending limits, approval gates, or safeguards. For a personal project that’s convenient. For anything with a real budget, you’d want guardrails before you let it loose on a checkout page.
Why It Matters
Most AI agents live entirely in software. Sprocket is trying to bridge into the physical world, from schematic to parts list to actual purchase. That’s a meaningful shift for makers, robotics tinkerers, and hardware startups who currently juggle CAD tools, procurement, and code across separate workflows. Whether the web-sourced reliability claim holds up in real builds is the open question, and early Hacker News users will be the ones to stress-test it. If you want the full technical details and install options, they’re in the original Show HN post.