Skip to main content

Module context_compression

engram::intelligence

Module context_compression 

Source
Expand description

Active Context Compression (RML-1211)

Provides adaptive, multi-level compression of memory content to fit within LLM context windows. Pure computation — no database access.

Compression levels:

  • None — full content, no changes
  • Light — remove stopwords and filler phrases
  • Medium — extractive summary (first sentence + entity sentences)
  • Heavy — key facts only (entities, numbers, dates)

Structs§

CompressedEntry
A memory entry after compression.
ContextCompressor
Adaptively compresses a set of memories to fit within a token budget.
MemoryInput
Simple memory representation used as compression input.
TokenBudget
A snapshot of token budget state.

Enums§

CompressionLevel
Level of compression to apply to memory content.