The smallest possible AI personal assistant for ESP32.
Zclaw is written in C and runs on ESP32 boards with a strict firmware budget target of <= 888 KB on the default build. It supports scheduled tasks, GPIO control, persistent memory, and custom tool composition through natural language.
Fun to use, fun to hack on.
I've put this on the "just get it" list because that ESP32-C3 costs like 10k sats incl shipping and it would be cool to show to all the people that bought whole mac minis just to pipe some json to an https OpenAI endpoint.
Yesss esp32 boards are very very simple to flash and use ! and very cheap... :)
I think I have half a dozen various versions laying around. Get chip to play with.
I think the first thing to do is swap out telegram for a private
ntfyendpoint, which will save compute and installing shit on a phone (telegram) that you don't want to have installed.From there on, offload all compute to a cheapo VPS and you can have it do anything, while you just carry the authorized software in your pocket, or if you wanna be a hipster, fanny pack.
Someone in Pleb Lab "reverse engineered" Signal and is creating a Slack-like interface for Signal to communicate with his bot child (also to communicate with frens cause he likes Slack's interface).
We've all been waiting for him to ship it. I'd love to get rid of Matrix for SN's internal comms.
If I were on Matrix I'd love to get rid of that too, haha
Lightest "signal" I've had is XMPP+OMEMO, but I think that that is still overkill for an ESP32 that also needs wifi and https. Which is why I said ntfy because you just run that over mTLS (current
mainbranch of the vendor apk supports it) and done.What's the purpose of this when the actual LLM isn't running on the ESP32 anyway?