bmad-distillator

Document distillation compresses multi-file source collections into a single, token-efficient context for LLM consumption. It removes human-oriented prose and noise while preserving every fact, decision, and constraint for downstream workflows.

Semantic splitting divides distilled documents into semantically coherent parts. An analyzer-driven routing pass evaluates your source files and recommends either single dense compression or fan-out semantic splitting when downstream LLM workflows require structured context.

Yes, lossless distillation supports completeness checks and optional round-trip reconstruction to prove fidelity. Validation ensures the compressed distillate preserves specific numbers, dates, rejected alternatives, and relationships from the original source documents.

Deduplication during distillation identifies and removes overlapping content across multi-file document collections. The analyzer groups related files and eliminates redundant information while maintaining every unique fact and constraint for LLM consumption.

Distillation works with multi-file collections like product briefs, discovery notes, research reports, and architecture docs. It produces a compact, machine-loadable distillate optimized for creating PRDs, designing architecture, or feeding agent pipelines.

Source documents require no special preparation before lossless LLM compression. The distillation process performs thematic analysis, compression, deduplication, and semantic splitting directly on messy, multi-file collections to produce a dense, machine-loadable context.