lex-cognitive-compression

Cognitive information compression for LegionIO. Models information compression and abstraction as knowledge moves through memory tiers.

Raw memories start with high fidelity but take full storage. Over time they compress — detail fades, gist remains. This extension models that process: chunks are stored at a compression ratio, can be incrementally compressed (losing fidelity), and can be abstracted into higher-level representations. Decompression is lossy — you cannot fully recover the original fidelity once lost.

Seven chunk types are supported: episodic, semantic, procedural, sensory, emotional, abstract, and relational. Each has different natural compression targets appropriate to its type.

client = Legion::Extensions::CognitiveCompression::Client.new

chunk = client.store_chunk(
  label: 'meeting notes 2026-03-14',
  chunk_type: :episodic,
  original_size: 1.0
)

client.compress_chunk(chunk_id: chunk[:chunk][:id], amount: 0.2)
client.average_fidelity
# => { fidelity: 0.96 }  # small fidelity loss from one compression step

# Abstract multiple chunks into a summary
client.abstract_chunks(
  chunk_ids: [chunk[:chunk][:id]],
  abstraction_label: 'Q1 meeting themes'
)

client.compression_report

bundle install
bundle exec rspec
bundle exec rubocop

MIT