I've built a source code graphRAG for Java/Kotlin project agentic analysis and development. It is built on top of jQAssistant with a detailed knowledge graphRAG in Neo4j.
You can check out the code here: https://github.com/2015xli/jqassistant-graph-rag
What it can do:
* "What's the main purpose of the com.example.auth package?"
* "Show me the call chain leading to the processPayment method."
* "What services use the UserRepository class?"
1. Graph Creation: It uses jQAssistant and Java/Kotlin source file parsers to analyze your code's structure, dependencies, and relationships. It essentially builds graph overlays for a source code tree and build artifact tree.
2. GraphRAG Enrichment: It then enriches the graph and generates summaries from individual methods and classes all the way up to packages and the entire project. Embeddings are generated for the summaries to facilitate semantic query.
3. MCP server and Agent: It exposes the graphRAG capabilities through an MCP server and an example coding agent. You can use them to accomplish complex tasks.
* Modular design that can be easily adapted to new graphRAGs for other languages.
* Parallelized summarization process and summary cache management to save the cost in money and time.
Btw, I also built a source code graphRAG for C/C++ development at https://github.com/2015xli/clangd-graph-rag.