Top AI Product

There’s something deeply satisfying about watching someone strip a problem down to its absolute bare bones. That’s exactly what [zclaw](https://github.com/tnm/zclaw) does — it takes the concept of a personal AI assistant and squeezes it into an 888KB firmware image running on an ESP32 microcontroller. The whole thing is written in C, and when you look at the size breakdown, the actual application logic is only about 35KB. The rest is Wi-Fi stack, TLS, and certificate bundles. That’s absurdly lean.

So what does it actually do? You talk to it through Telegram or a web relay, and it can control GPIO pins, run scheduled tasks with timezone awareness, remember things across reboots, and chain tools together — all through natural language. It supports Anthropic, OpenAI, and OpenRouter as LLM backends, so you’re not locked into one provider. The recommended starter board is a Seeed XIAO ESP32-C3, which you can grab for around five bucks. It also runs on the ESP32-S3 and C6.

The project [hit the Hacker News front page](https://news.ycombinator.com/item?id=47100232) on February 22nd, pulling in over 220 points and 117 comments. The discussion was predictably lively — some folks were genuinely impressed by the engineering discipline required to hit that 888KB target, while others questioned whether this is just “an LLM in a loop calling APIs.” Fair point, but I think the skeptics are missing the bigger picture here. The always-on, zero-maintenance angle is real. No Linux to patch, no Docker containers to babysit, no server bills. Just a tiny chip drawing milliwatts, quietly doing its thing.

The [field guide](https://zclaw.dev/) has solid documentation if you want to get started. Flash it, provision your Wi-Fi credentials and API key, and you’re up. The whole setup process is surprisingly smooth for a bare-metal project. If you’ve been looking for a weekend hardware project that actually does something useful after you’re done soldering, zclaw is worth a look.