bmad-distillator

Lossless distillation compresses source documents by removing human-centric overhead while preserving every fact, decision, and relationship. This produces a dense, LLM-ready artifact with token estimates for downstream workflows like PRD creation or architecture design.

Packaging large documents into a single context involves analyzing inputs, compressing groups, and verifying completeness. You can generate a portable distillate or semantically partitioned parts, optionally specifying a downstream consumer and token budget to fit your pipeline.

Semantic splitting for downstream LLM pipelines is achieved through fan-out routing, which partitions a distillate into semantically coherent parts. This preserves all factual content and constraints while fitting specified token budgets for tasks like long-form content synthesis.

Yes, you can validate document compression by running an optional round-trip validation step. This high-stakes verification follows the analyze, compress, and verify activation steps to ensure the distillate is completely lossless.

No external dependencies are required to perform lossless document distillation. The tool operates self-contained using its internal scripts to analyze inputs, compress groups, verify completeness, and output portable distillates with token estimates.

You should avoid lossless distillation if your downstream LLM workflow requires preserving the original document's narrative structure or human-centric formatting, as the process specifically strips human overhead to produce a dense, factual artifact.