bmad-distillator

Lossless document distillation compresses source files into token-efficient outputs by applying deduplication and semantic splitting while preserving every fact and relationship. Optional round-trip validation confirms no information was dropped during compression.

Semantic splitting divides large distilled documents into a structured folder of outputs based on token budgets, allowing downstream agent pipelines to process specific context segments without exceeding model input limits.

Yes, document distillation accepts folder paths and glob patterns as input, routing multiple source files through analyze-then-compress logic to produce a single dense distillate for downstream consumers.

Round-trip validation checks losslessness by attempting to reconstruct the original source information from the compressed distillate, ensuring completeness when the --validate flag is set during processing.

While semantic splitting helps manage strict token budgets, extremely large document sets may require processing in batches, and distillation effectiveness depends on the quality of input deduplication and compression rule application.