Over the past few months our team has been building the Kiraa Engine and the missing software framework that lets Apple silicon do what it’s actually capable of. Apple has the best hardware on the market, but the software gap has been holding it back. So we wrote hundreds of thousands of lines of optimised code to close it.
Now we’re releasing two projects as open source, and inviting you to build with them.
SwiftPandas is a native port of Python’s pandas, with performance uplifts of up to 80x. Jobs that used to take days now take hours.
Project Onyx is a lightweight LLM that runs entirely on device. It proves you can get solid output quality and fast response times without ever touching the cloud.
We’re opening these up because the more developers building on device, the faster the world realises that cloud AI is a bad idea. Expensive, risky, and unnecessary when fast, cheap and secure models can run right on your own hardware.
If you’re building something interesting with SwiftPandas or Project Onyx, I want to see it. Accelerated an old pandas workflow?