bmad-distillator

Document distillation compresses multiple source documents into a single or semantically split distillate that preserves every fact and relationship while removing redundancies for token-efficient downstream AI workflows. You provide input paths and optionally use a --validate flag.

Lossless document distillation is the process of compressing information into a distillate that preserves every fact, decision, constraint, and relationship. It optionally performs round-trip validation to reconstruct originals from the distillate to verify no data was lost.

Yes, semantic splitting is a core feature of document distillation. You can compress information into a distillate with optional semantic splitting, allowing you to generate sections specifically for targeted analysis, design, and decision pipelines.

The best way to verify lossless compression is using the optional round-trip validation feature. It reconstructs originals from the distillate to verify format and confirm that every fact, decision, and relationship was preserved during the document distillation process.

You do not need to add metadata manually because the distillation process automatically includes frontmatter metadata for provenance. This ensures deterministic AI pipelines can trace compressed distillates back to their original source documents.

Large documents exceed token limits and cause context loss in LLM workflows. Document distillation solves this by analyzing input sources, determining grouping and splitting strategy, and compressing information into token-efficient, context-rich distillates with frontmatter metadata.