skillgraph
open-source agentic ai framework that reduces llm costs by 89%
skillgraph is an open-source agentic ai framework (apache 2.0) that dramatically reduces llm operational costs through intelligent prompt caching and skill-based architecture.
the problem
most ai applications today rely on complex rag (retrieval-augmented generation) systems that are expensive to run and difficult to maintain. every query hits the llm, even for similar questions that have been answered before.
the solution
skillgraph replaces complex rag systems with a subject-object memory architecture. instead of retrieving and processing documents for every query, it learns and caches skills-reusable patterns of reasoning that can be applied to new situations.
key features:
- 89% cost reduction through intelligent prompt caching
- skill-based architecture that learns and improves over time
- subject-object memory for efficient knowledge representation
- drop-in replacement for existing llm workflows
technical approach
the framework introduces a novel way of thinking about ai agents. rather than treating each interaction as a fresh context window, skillgraph maintains a graph of learned skills that can be composed and reused.
this means your ai gets smarter and cheaper to run the more you use it.
current status
actively maintained and in production use. contributions welcome.