All notable changes to this project will be documented in this file.
- Function-Level Chunking: GDScript files now parsed semantically - each
func,signal, and@exportbecomes a separate searchable unit with rich metadata instead of basic line-based chunks - Signal Flow Tracking: Complete understanding of signal emission → connection → handler chains across entire projects
- Multi-Hop Dependency Tracing: Trace function call chains across files (e.g., Input → Controller → Physics → Animation) with configurable depth
- Enhanced Search Modes:
- Semantic search (AI understanding of code intent)
- Keyword search (exact text matching with proper BM25 implementation)
- Hybrid search (combines both approaches)
- Smart Search Intelligence: Auto-detection of search intent, cross-section deduplication, exact match verification
- Graph Centrality Analysis: Identify architecturally important files and their structural roles in projects
- Performance Improvements: ~2.7x more intelligent chunks from same files (51 → 138 chunks in demo project)
- Migrated from BigQuery to Weaviate: Enhanced performance and better vector search capabilities
- Three Collection Architecture:
ProjectEmbedding: Function-level chunks with dependency metadataProjectGraph: File-to-file relationshipsProjectDependencies: Function-to-function dependency mapping
- Real-Time Updates: All collections updated automatically when files change, including dependency re-extraction
- Advanced Filtering: Eliminates indexing of
.importfiles and other binary assets - Weaviate v4 Compatibility: Updated client implementation with proper error handling and fallback strategies
- Dependency Types Tracked: Function calls, signal emissions, node access, physics API usage, input system access
- Context-Aware Results: Search results include function metadata (signals emitted, functions called, nodes accessed)
- Nuclear Filtering System: Multi-layer filtering prevents irrelevant results (
.importfiles, etc.) - Exact Match Verification: Keyword search validates actual text presence in content
- Enhanced Error Recovery: Graceful fallbacks when Weaviate BM25 API encounters issues
- New Technical Documentation: Added comprehensive
backend/indexing.mdwith architecture details, examples, and troubleshooting - Updated Backend README: Modernized
backend/README.mdto reflect Weaviate-based system with enhanced API documentation - Enhanced Main README: Added Advanced Indexing System section and backend documentation links throughout
- Environment Configuration: Updated setup instructions for Weaviate requirements (
WEAVIATE_URL,WEAVIATE_API_KEY)
- Enhanced
/search_across_projectendpoint: Addedtrace_dependencies,search_modeparameters - Improved Response Structure: Results now include
chunk_type,function_name,dependencies,signals_emittedmetadata - Debug Logging: Comprehensive logging for function extraction, dependency tracking, and search execution
- Performance Metrics: Detailed timing and chunk count reporting
- Tightened file management tools: apply-edit now outputs structured diff JSON instead of code (1e8ede8e55).
- updates sprite and sprite sheets
- Apply-edit: create structured diffs; agent can view live and debugger errors (6626e21ec5).
- Updated
.gcloudignore(31cc62bd4d). - Indexing and search: improved graph creation and querying; integrated latest Godot docs indexing as a tool (dd05c8b4e5).
- Misc: updated contact email (17ee1973c0).
- Apply-edit: ability to apply patches at specific file locations (390e3b530f).
- Apply-edit: text updates (7830f90f2e).
- README: added Discord link (d2d77e92ca); typo fix; general updates (b91198bcd6, d88c5f0296).
- Editor: script editor whitespace cleanup (de0a870f6e).
- Initial import (clean history; removed sensitive data) (abab192012).