Function that calls an LLM with (system, user) prompts.
Optional traits: HexacoTraitsOptional HEXACO personality traits for bias-aware compression.
Compress a batch of observation notes into denser summaries.
The method:
Batch of observation notes to compress.
Array of compressed observations. Returns empty array on LLM failure.
LLM-based compressor that takes a batch of ObservationNote objects and produces denser CompressedObservation summaries.
Achieves 3-10x compression while preserving key facts, entities, and temporal context. Each compressed observation carries three-date temporal metadata: when the compression happened, the earliest referenced event, and a human-friendly relative time label.