bmad-distillator

You can compress large documents for LLM context optimization by distilling them into a bullet-only format. This lossless compression preserves all decisions, constraints, and relationships while removing prose overhead for downstream workflows.

Semantic splitting for token budgeting divides source documents into section-specific distillates within a split-folder layout. It analyzes sources to route content into manageable chunks that fit target token limits while maintaining structural integrity.

Round-trip validation verifies lossless document compression by using a reconstructor subagent to rebuild the original sources from the distilled output. It then reports any gaps or hallucinations to ensure factual completeness.

Yes, you can distill multiple source documents into a single LLM-optimized distillate. The router-driven workflow analyzes the sources to determine whether to produce one unified file or semantically split sections.

The bullets-only format for information extraction removes all narrative prose, which limits readability for humans. It is strictly designed for LLM context optimization and requires YAML frontmatter to structure the extracted facts and themes.

You need document distillation for LLM workflows when human-oriented source documents are too large for target token budgets. It hyper-compresses text into structured facts and decisions while preparing the output for downstream consumption.