feat:Implement Hybrid Search with Call Graph for Better Context Connectivity #33
Description
Problem
Currently, amdb relies solely on vector similarity search. This leads to context fragmentation, where the AI misses logically connected code (e.g., interface implementations, function callees, parent classes) if they don't share similar keywords. The "God Mode" promise is limited by this lack of structural understanding.
##Proposed Solution
- Enhance src/core/graph.rs to build a lightweight Call Graph or Symbol Dependency Map during the scanning phase.
- Modify the search logic (src/core/vector_store.rs) to apply a Hybrid Search algorithm:
- Step 1: Perform Vector Search to find the top-k relevant chunks.
- Step 2: Traverse the Graph to find direct dependencies (callees/callers) of those chunks.
- Step 3: Rerank or include these connected chunks in the final context.
Definition of Done
- Graph struct implemented to store file/function relationships.
- Parser updated to extract basic import/call relationships (starting with Rust/TS).
- Search query returns not just semantically similar files, but also their dependencies.
- Integration test added: Searching for a function definition also returns the file where it is heavily used.