vitali87/code-graph-rag v0.0.100 on GitHub

Highlights

Performance: Parallel flushing, asynchronous MCP handlers and periodic file flushing enhance data handling speed.
Reliability: Improved data reliability through enhanced flushing mechanisms.
Efficiency: Optimised data management with parallelisation of key operations.
Architecture: Introduction of asynchronous MCP handlers for improved responsiveness.

Validate LLM-generated Cypher queries against destructive keywords with regex word boundaries
Block comment-based injection bypasses (LOAD/**/CSV, LOAD //\nCSV)
Flush graph nodes and relationships in parallel using a thread pool with per-thread connections
Run all MCP handlers async via asyncio.to_thread to prevent event loop stalls
Batch Qdrant upserts (default 50) with retry and exponential backoff
Clean up Qdrant embeddings when deleting a project
Reconcile expected vs stored embedding IDs after indexing and log any gaps
Protect the database connection with threading.Lock and serialize MCP writes with asyncio.Lock
Add validated config settings for retry counts, batch sizes, flush intervals, and thread pool size
Add 72 new tests for Cypher validation, vector store operations, reconciliation, and MCP helpers

Full Changelog: v0.0.99...v0.0.100

perf: add parallel flushing, async MCP handlers, periodic file flush, and data reliability by @vitali87 in #389

Full Changelog: v0.0.99...v0.0.100