bmad-distillator

To compress documents for LLM consumption without losing facts, apply a lossless distillation process that analyzes, deduplicates, and extracts facts, decisions, and constraints into a single optimized output. This preserves all critical relationships while minimizing context overhead.

To semantically split a distillate output, configure the compression process to generate multi-part outputs with section manifests. This breaks the single compressed document into logical semantic sections, optimizing context delivery for downstream LLM workloads.

Yes, you can process mixed document types like product briefs, discovery notes, architecture docs, and research reports together. The distillation pipeline handles facts, decisions, constraints, open questions, and named entities across diverse document formats.

Round-trip validation during document distillation verifies that the compressed distillate accurately preserves information from the source documents. This optional step follows the analysis, compression, and verification pipeline to ensure lossless information extraction.

The best way to reduce token overhead when feeding multiple reports to an LLM is lossless document distillation. It converts multiple source documents into a single compressed distillate, applying deduplication and rule-based compression to maximize token efficiency without data loss.