CacheLane is a local prompt-caching and trajectory-aware orchestration layer. It segregates your context into volatility tiers and prunes stale tool outputs, keeping your sessions fast and affordable.
# Install globally
npm install -g cachelane# Integrate with Claude Code hooks
cachelane install# Diagnose and verify local setup
cachelane doctorSee how standard conversation context cost compounds vs. CacheLane's region reordering and K-pruning.
No breakpoint management or context pruning.
Orchestration reorders blocks; K-pruning stubs idle content.
CacheLane operates completely locally behind the scenes, ensuring optimal pricing without manual intervention.
Automatically classifies incoming context blocks and aligns them into three regions with dual `cache_control` breakpoints.
Identifies stale files and tool-outputs unreferenced for consecutive turns and swaps them out for light, refetchable stubs.
Schedules low-cost, single-token background pings when idle to keep your prompt cache hot, avoiding cold-start write penalties.
All references, database metrics, and logs are kept in a local SQLite file. Your prompts and files never leave your environment.
Built to be completely invisible. Any internal runtime error gracefully returns the raw prompt, never breaking your editor session.
Use simple command line options like `cachelane stats` or `cachelane explain` to get transparent reports of your cache savings.