Skip to main content

Vismriti Engine

The intelligent compression engine at the heart of Vidurai.

Overview​

Vismriti (Sanskrit: विस्मृति, "strategic forgetting") is Vidurai's compression engine that intelligently manages memory by selectively compressing or discarding less important information.

How It Works​

  1. Monitors token usage across all Koshas
  2. Identifies compression candidates based on importance scores
  3. Applies LLM-based semantic compression
  4. Preserves high-importance memories

Compression Strategies​

Conservative​

  • Compress only when absolutely necessary
  • Higher quality preservation
  • Higher token costs

Balanced (Default)​

  • Smart compression when approaching limits
  • Good balance of quality and cost

Aggressive​

  • Proactive compression
  • Lower token costs
  • May lose some detail

Configuration​

from vidurai import Vidurai

# Balanced approach (default)
memory = Vidurai()

# Aggressive compression
memory = Vidurai(compression_threshold=0.7)

# Conservative approach
memory = Vidurai(compression_threshold=0.9)

See Compression Strategies for advanced usage.