bmad-distillator

Losslessly compress documents for LLM context by distilling them into token-efficient, bullet-only output that preserves every fact, decision, and relationship without summarization. This semantic compression strips overhead while maintaining complete information integrity.

Semantic splitting for token budgets automatically divides large distillates into self-contained sections with a root orientation index when token limits are exceeded. This ensures the compressed output remains within LLM context windows while preserving all extracted information.

Extract information from multiple source documents by running source analysis for routing and splitting, then producing thematically grouped bullet-only output with headings. This creates a structured distillate suitable as sole LLM context for downstream PRD, architecture, or generation tasks.

Verify losslessness through optional round-trip validation that reconstructs original documents from the distillate when subagent support is available. Additionally, structured completeness and format checks run automatically with targeted fix passes if any gaps are detected.

Reduce token overhead by applying semantic compression that strips human and LLM overhead while maintaining completeness. The process produces hyper-compressed distillates with thematically grouped bullets, returning structured JSON results for easy integration into LLM workflows.