AI agents

Prompt caching and cache savings

• Prompt Caching: How It Works and When It Pays
• A Cache Breakpoint Lint Rule for Prompt Templates
• A TTL Matrix for Two Cache Layers
• Cache Hits vs Prompt Freshness: The Trade-off
• Cache Invalidation Across Two Layers: A TTL Matrix
• Comparing Prefix Caching Rules Across Three Providers
• Eviction Policy for a Two-Layer Cache: LRU, LFU or Length
• Stacking Local Cache on Provider Prefix Cache
• The Invalidation Event Contract: Fields a Publisher Must Send
• A Local Cache Layer Under Provider Prefix Caching
• A Stale Hit That Reached a Customer: Reproducing It From Logs
• Cache Hit Rate Is Not Savings: A Cost Model per Layer
• Cache Keys That Survive a Tenant or Prompt Version Change
• Cache Warming Across Two Layers: What to Preload and When
• Cache Warming and Eviction: Keeping the Hot Prefix Resident
• Caching Memory Reads: Keying, TTL and When a Cached Row Is Wrong
• Detecting a Cache Hit That Returned Wrong Context
• Diagnosing a Cache Hit That Returned Wrong Context
• Estimating Cache Savings Before You Restructure a Prompt
• Fine-Tuning or Prompt Caching: How to Choose
• Hit Rate per Feature: What to Log
• Hit Rate, Stale Rate and Cost Saved
• Keying a Cache on Tenant and Prompt Version
• Keying a Cache on Tenant and Prompt Version Fields
• Logging Cache Hits, Tokens and Latency per Call
• Logging Prefix Identity per Call: A Field Guide
• Logging Which Cache Layer Served a Call: A Schema
• Logging Which Cache Layer Served a Call: A Schema and a Join
• Reading a Provider's Cache Telemetry Without a Dashboard
• Replaying a Day of Traffic With Caching Off
• Semantic Caches in Customer Support
• Staleness Budgets per Data Source
• Testing a Cache Stack: Fixtures for Hits, Misses and Invalidation
• Trimming Prompts and Context to Cut Input Tokens
• Version Tokens That Roll Back a Cache
• Warming a Cache From Traffic: What to Preload and What to Skip
• When a Cache Hit Is Staler Than the Tools It Already Read
• When a Cache Hit Returns the Wrong Context
• When a Provider Deprecates Prefix Caching: A Fallback Plan
• When TTL and Eviction Disagree: A Debugging Walkthrough