bmad-distillator

Document distillation compresses source materials into hyper-compressed, token-efficient distillates that preserve every fact and relationship while removing overhead. This process prepares lossless, LLM-optimized context files for downstream workflows like PRD creation or architecture design.

Yes, semantic splitting with token budget control automatically switches from a single dense distillate to a multi-part folder when content exceeds target thresholds. This approach manages token budgets by dividing distillates semantically by size for downstream agent pipelines.

Lossless extraction verifies distillates through completeness and format checks, and can optionally perform round-trip reconstruction validation. This semantic diff process reconstructs sources from the distillate to test whether all original facts and relationships are preserved.

The best way to prepare multiple documents for token-budgeted pipelines is distillation, which runs analysis, compress, and verify stages to produce dense, LLM-ready context files. This ensures you maximize token throughput without losing critical constraints or decisions.

You should use document distillation when wasting context budget and token throughput is a concern, or when you need to feed multiple source documents into an LLM without losing facts. It is essential for information-extraction scenarios requiring strict token budget control.

No, the distillation process requires no external dependencies, relying entirely on its internal scripts, references, and assets. You only need to provide the caller-provided source documents and an output path to generate the lossless distillate.