Knowledge Compilerconcept

The Knowledge Compiler is the original project that turns collections of Markdown documents into statically-deployable semantic artifacts (knowledge graphs, concept hierarchies, vector embeddings, cluster maps) via a multi-pass compilation pipeline — no runtime LLM inference required.

It asks why an AI system should re-discover the same knowledge on every query, and answers by amortizing semantic analysis into a build step that emits a self-contained reasoning substrate.

Where it appears

Descend

Historical evolution

The first public statement of the pattern (2026-07-11): 9-phase pipeline, 113 docs -> 192k concept nodes in 1.5s.

Implementation details

Monorepo of 7 packages + web app; phases: parsing, analysis, graph, embedding, clustering, optimization, generation.

Sources

View in graph →