LoreTokens
279:1 lossless · Wikipedia · 75s
A post-Shannon semantic primitive. Compressed form stays addressable - no decode-through-a-model overhead. Tested against a 1 GB Wikipedia dump (166,565 articles) at L8.
Primitives
Each of these was built because the available tooling couldn't do the job. Each has a working reference implementation and a Zenodo preprint filed as prior art.
SAIQL
30-80× query compression vs. Postgres
Semantic AI Query Language. Standalone, zero dependencies, pip-installable. The query layer over compressed semantic state.
Atlas
First deterministic semantic RAG
Same query, same records, every time - with a cryptographic proof bundle showing which records were consulted and why. Retrieval that can reproduce its own reasoning.
QIPI
Exact-lookup index
The index that sits under Atlas. Reproducible retrieval at vector-search speed. Built for compressed semantic tokens, not embeddings.