Project Description: Library Hyperboria, Cloud Runemacs Thesis: Editing, Archiving, and Indexing text is the primary use case for computers, most (not all) of the challenges there come from systemic disorder, not essential technical complexity. Goal: Fully self hosted, offline first, knowledge database, library, and communication platform. Feature: Excellent text editor in PLAN, fully self hosted. First class emacs/vim/kak bindings (can support all at once fairly seamlessly using modes) Feature: no Unicode, 8bit text, monospace, European languages only. (Massive decrease in difficulty) Feature: p2p messaging, mail, and publishing. Integrated directly into the editor. (New protocols) Feature: all content published via a network of p2p swarms / merkel dag. Like IPFS/BitTorrent, but PLAN native. Featue: no web, SSL, HTTP, SMTP, SMS, DNS, etc. (massive decrease in difficulty). Feature: Offline Global search index (full text over global corpus) Feature: fully integrated local LLM to achieve "talking library" dynamic. Feature: specific targeted Hardware/Software bundling (Strix Halo, etc) A very basic version of this is realizable fairly easily, expand capabilities slowly over a long timespan.

LLM summary of this character set sketch: An 8-bit linguistic encoding for modern European text, prioritizing strong support for Germanic (English, German, Swedish, Norwegian, Icelandic) and Romance (French, Spanish, Portuguese) languages, plus Greek and Russian as neighboring scripts with major literary traditions. Space is saved primarily through cross-script glyph unification (shared forms like A/Α/А collapse to single code points), a global "no accented capitals" rule (caps-accents being conventionally droppable in Greek, French, and Italian anyway), and the rejection of non-linguistic content — math symbols, currency variants, editorial marks, and curly quotes all cut in favor of $, --, ..., and bare letters. ASCII compatibility was abandoned to free the lower 128 slots, and hard monospace was preserved by forbidding combining marks and escape codes.

The cog stuff from earlier architectures is no longer a part of PLAN. It has been replaced with direct support for side-effects, with virtualization to run code in a pure sandbox.

To be specific, the plan is to have something akin to resumable exceptions. I am confident that this will work, but exactly how much expressive power can be made available needs to be evaluated w.r.t performance concerns. A spectrum of designs exists.