bmad-distillator

Distillation compresses multi-file markdown, text, YAML, and JSON sources into dense thematic bullets, preserving facts and decisions while stripping filler to reduce token usage. It applies thematic extraction and deduplication to create a high-signal, token-efficient distillate.

Lossless document distillation extracts key information from source documents into dense thematic bullets while preserving named entities, numbers, decisions, rejected alternatives, and dependencies. It ensures no critical facts are lost during compression, making it ideal for research synthesis and PRD creation.

Semantic splitting handles large inputs by producing a root index plus self-contained section distillates when token budgets demand it. This approach divides oversized documents into manageable sections, ensuring each distillate remains self-contained and contextually coherent.

Completeness validation checks the distillate against source headings and named entities to reduce information loss. Optional round-trip validation further verifies that all critical facts, decisions, and constraints from the original documents are preserved in the compressed output.

Yes, distillation reads and compresses markdown, text, YAML, and JSON source sets together into dense thematic bullets. This multi-format support allows architecture design workflows to consolidate structured configuration data and unstructured documentation into a single lossless distillate.

Document compression for PRD creation preserves named entities, numbers, decisions, rejected alternatives, scope boundaries, open questions, risks, and dependencies. It strips filler content while retaining all critical project requirements and constraints needed for downstream LLM workflows.