bmad-distillator

Document distillation compresses multiple source documents into a compact, token-efficient distillate file. The process preserves facts, decisions, and relationships while routing and semantically splitting content across files to maintain cross-document coherence.

Lossless distillation is the compression of source documents into compact outputs without losing facts, decisions, or constraints. It is needed when preparing large product briefs, architecture docs, or research reports as single consumable inputs for downstream LLM workflows.

Yes, distillation supports multi-file processing with semantic splitting and routing. It maintains cross-document coherence across PRDs, architecture docs, and discovery notes, ensuring relationships and constraints are preserved in the final distillate.

You can specify a token budget as an optional parameter when running the distillation process. The distillator uses this budget to compress the source documents list into a distillate with frontmatter and structured bullets tailored for the downstream consumer.

Distilling large research reports into a single compact distillate is the best approach for LLM consumption. This lossless compression outputs a file with frontmatter and structured bullets, preserving complete context while minimizing tokens.

Distillation works with both single and multi-file source sets. You provide a source documents list at an input path, and the distillator processes it into a single distillate file, optionally allowing you to specify an output path and token budget.