Skip to main content

Context Algorithms

MemexLLM uses context algorithms to manage how conversation history is presented to LLMs. These algorithms ensure optimal context while respecting token limits.

Available Algorithms​

FIFO Algorithm​

Coming Soon​

We're working on additional context algorithms:

  • Select messages based on relevance
  • Use embeddings for similarity

Token-Aware​

  • Smart token counting
  • Model-specific optimizations

Time-Based​

  • Select messages by recency
  • Configurable time windows

Hybrid​

  • Combine multiple algorithms
  • Custom selection strategies

Algorithm Features​

All context algorithms in MemexLLM provide:

  • Message Selection: Choose which messages to include in context
  • Token Management: Respect LLM token limits
  • Thread Support: Work with conversation threads
  • Metadata Support: Use metadata for selection
  • Type Safety: Full type hints support

Custom Algorithms​

Need a different selection strategy? You can create your own:

Next Steps​