Kynd Agent / Blog

Notes from the edge.

Field notes from building a fully local AI companion on Apple Silicon — what shipped, what we measured, and how to run it on your own Mac. Everything here happened on real hardware, most of it on an M1.

Latest

An animated portrait of Kynd’s maker in a golden theatrical room

Community · earnings calculatorYour Mac has more to giveExplore what your idle Mac could earn, count the electricity, and see the films a community of Macs could help create. We are just beginning.

A filmstrip of scenes made in Kynd Agent, all rendered locally on Apple Silicon

Coming soonThe cloud, without the data centreKynd Share turns the Macs people already own and rarely use into rendering capacity—and pays the person whose Mac did the work.

A detailed red fox generated locally at 1024 by 1024 through Kynd’s streamed Z-Image engine

New updateYour RAM shouldn’t decide what you can createKynd’s streamed Z-Image engine gives 16 GB and 64 GB Macs the same model, the same creative canvas and the closest possible quality.

A claymation musician performing in a camper van during a multi-shot LTX 2.5 take

VideoTwenty seconds. Four shots. Now on a 16 GB Mac.The full LTX 2.5 Social experience—multi-shot video, real vocals and continuous lip sync—now runs from 16 GB upward.

A claymation busker singing in his camper van — rendered locally by the streamed LTX 2.5 engine

VideoLTX 2.5 now fits a 16 GB MacThe 32 GB floor falls: the streamed engine renders a five-second Ingredients clip in a measured 13.3 GB peak — with checksum-identical output.

Frames from an LTX 2.5 Ingredients render with a consistent character

ReleaseTwo local leaps, shipping togetherLTX 2.5 Ingredients gives local filmmakers a reusable cast. Qwen 3.8 27B gives local agents a formidable new brain.

Kynd's Build mode running a local agent team on one Mac

AgentsOne Mac. A whole agent team.A main builder, responsive conversation and background specialists — no cloud hand-off, every model local.

Kynd's calibration screen benchmarking a Mac

BenchmarksHow to make Qwen 3.8 27B faster on an M1 MacThe adapters with the loudest speedup claims made the model slower on M1. The data, and the tuning that gave 1.48x instead.

Run it yourself

A claymation character reference used to condition an LTX render

GuideHow to run LTX 2.3 & 2.5 on MacThe 16–64 GB tiers, model sizes, resolutions and Ingredients support — which build fits your Mac.

Kynd chatting with a local Qwen model on a Mac

GuideHow to run Qwen 3.8 27B on MacUnified memory and quant size decide whether it fits — and how much useful context remains.

GuideHow to run Z-Image Turbo on MacA 6B image model, distilled for speed, on Apple Silicon through MLX — with a practical 16 GB route.

Start free. Keep it free.

No account, no card. Apple Silicon Mac with 16 GB or more.

Download Kynd free