The latest update is 25.12. It's great.
This release brings meaningful refinements to LatteAI — speed, polish, and an updated app icon to tie it all together.
December, 22 2025
LatteAI — Now Even Faster ☕
We've made significant performance improvements across the board, with particular gains on the latest M4 Macs.
The LatteAI interface has been refined:
- Edit Popup — Streamlined and easier to use. Making inline changes with ⌘; now feels even more natural.
- Chat View — Polished for clarity and smoother interaction.
Improved Auto-Complete
The auto-complete popup interface has been improved to make it easier to use and provide more information.
And…
- Fixed a crashing bug when syncing files.
- Fixed a crashing bug when searching for files.
- Improved a drawing glitch in the navigator.
- Updated application icon.
But underneath its rubbery keyboard and distinctive rainbow stripe lies a feat of minimalist engineering that still teaches lessons to modern hardware designers. At the heart of the machine lies a single, mysterious chip: the .
The Spectrum’s ULA implements a non-transparent memory access. The Z80 runs at 3.5 MHz, but the ULA reads video memory at 7 MHz during active scanlines. When the Z80 tries to access the same address range ($4000–$7FFF), the ULA:
Think of a ULA as a breadboard of unconnected NAND and NOR gates. You, the designer, pay for a metal mask that connects these gates into whatever logic function you need. It is a semi-custom ASIC. For a low-volume product (relative to Commodore), it was perfect.
In 1982, Sinclair Research released the ZX Spectrum, a machine that would define a generation of programmers and gamers. At its heart was not a standard chipset, but a single piece of custom silicon: the Uncommitted Logic Array (ULA) designated 5C112E. This paper argues that the ULA is not merely a peripheral controller but the philosophical manifesto of Sir Clive Sinclair. By analyzing the ULA’s dual role as video generator, DRAM multiplexer, and I/O traffic cop, we deconstruct the extreme cost-reduction strategies that birthed the home computing revolution. We will explore how the ULA’s infamous "contention" (the slowdown of the CPU to draw the screen) is actually a brilliant systems integration hack, and how modern FPGA recreations (like the Harlequin project) reveal the original designer’s trade-offs between component count and logical perfection.
: Documentation of how the ULA generates video signals, including deviations from standard PAL sync signals.