Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
Tuple<Query, List<CommunitySummary>> -> GlobalSynthesizedAnswer
Generate parallel intermediate answers from individual community reports, then aggregate them with a final LLM reduce step to synthesize a global answer.
Problem it solves
Answering holistic queries across an entire corpus exceeds the context window of standard single-turn LLM generation.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.