token-reduction

[EMNLP 2024 & AAAI 2026] A powerful toolkit for compressing large models including LLMs, VLMs, and video generative models.

~97% token reduction for AI coding sessions — zero deps, 33 languages, MCP server

Save tokens. Maximize context, Safely

AI-powered text compression library for RAG systems and API calls. Reduce token usage by up to 50-60% while preserving semantic meaning with advanced compression strategies.

A deterministic, zero-dependency prompt-pruning layer for long-running LLM conversations — expires stale tool state, collapses duplicate context, and proves it never drops a fact a later turn still depends on.

The official implementation of CVPR Workshop 2025 paper: Window Token Concatenation for Efficient Visual Large Language Models.