Product essays, launch updates, and implementation notes from the team building LATCH: document memory at the model level, shipped as portable .latch and .latchdoc files.
Starting point: the first post explains what LATCH is, why standard context-window workflows break down, and why the product is being built around compiled, persistent memory instead.
The blog starts with the launch post and can expand into product updates, benchmark breakdowns, architecture notes, and customer-facing explainers without leaving the current site language.
Why repeated context-window reading is a structural dead end, where RAG breaks on cross-document reasoning, how LATCH compiles a corpus once into persistent neural memory, and what the measured speed and cost numbers look like on the flagship Qwen 2.5 14B runtime.
The blog explains the thesis. The documentation and self-hosted evaluation license show the current product surface in detail.