bmad-distillator

Lossless document distillation compresses multiple source documents into a dense, token-efficient format. It analyzes mixed sources like markdown, txt, yaml, and json, then outputs a single distillate or semantically split set while preserving all facts and relationships.

Yes, you can specify an optional token_budget to constrain the output size for downstream LLM workflows. This ensures the generated distillate fits within your model's context window while retaining all critical facts, decisions, and constraints.

Lossless document distillation compresses source documents into a dense format without losing facts, decisions, or relationships. You need it when preparing large mixed-format knowledge bases for efficient downstream LLM consumption.

You validate distillate semantics by using the round-trip validation stage. Triggered by a flag, this stage verifies the compressed output by optionally reconstructing it to ensure all original facts and relationships are preserved.

Yes, distillation works with mixed source files including json, yaml, txt, and markdown. It analyzes these varied formats and compresses them into a unified distillate or semantically split outputs for downstream LLM workflows.