How does it calculate cache effectiveness?

It compares actual spent costs against the estimated cost of a full LLM generation to provide a 'Cost Saved' metric and hit-rate percentage.

Can I track costs for different agents separately?

Absolutely. It uses metadata attribution to link costs to specific 'agent_type' tags within a multi-agent hierarchy for precise billing.

Does it support provider-native prompt caching?

Yes, it specifically monitors 'cache_read_input_tokens' to account for and report on savings from provider-level prompt caching.

Does this skill require an external service?

Yes, this skill is designed to work with Langfuse for observability, metadata storage, and cost tracking calculation.

LLM Cache Cost Tracking

Name: LLM Cache Cost Tracking
Author: yonatangross

byyonatangross

•

Analytics & Monitoring

Monitors LLM expenditures and analyzes cache efficiency using Langfuse observability for complex multi-agent systems.

About

The Cache Cost Tracking skill provides a robust framework for monitoring LLM expenses and evaluating cache performance within AI applications. By leveraging Langfuse integration, it allows developers to track costs across multiple cache layers, attribute spending to specific agents in multi-agent workflows, and calculate real-world financial savings from prompt and response caching. This skill is essential for production-grade systems where token usage optimization and detailed usage reporting are required for budget management.

Key Features

Multi-layer cache hit tracking (L1, L2, Semantic, and Provider-level)
Customizable cost queries for historical and per-agent analysis
Hierarchical cost rollup for complex multi-agent workflows
Automatic token and cost attribution via Langfuse decorators
29 GitHub stars
Automated reporting on cache hit rates and estimated cost savings

Use Cases

Attributing LLM costs to specific agents or features in a multi-tenant production environment
Generating automated financial reports for AI token usage and budget forecasting
Auditing the ROI of implementing semantic caching versus direct LLM API calls

About

Key Features

Multi-layer cache hit tracking (L1, L2, Semantic, and Provider-level)
Customizable cost queries for historical and per-agent analysis
Hierarchical cost rollup for complex multi-agent workflows
Automatic token and cost attribution via Langfuse decorators
29 GitHub stars
Automated reporting on cache hit rates and estimated cost savings

Use Cases

Attributing LLM costs to specific agents or features in a multi-tenant production environment
Generating automated financial reports for AI token usage and budget forecasting
Auditing the ROI of implementing semantic caching versus direct LLM API calls