vector-store-workflow

To set up semantic code search with Qdrant, configure an embedding provider, create a Qdrant collection, index your repository using AST chunking, and execute filtered queries with re-ranking for relevant results.

AST-based chunking breaks down source code into functions, components, and types to create precise embedding vectors. This granularity improves semantic search accuracy by matching structural code boundaries rather than arbitrary text splits.

Yes, the workflow supports configuring multiple embedding providers including Ollama and OpenRouter. A pluggable provider facade abstracts the connection, allowing you to switch providers while maintaining the same indexing pipeline.

Incremental updates process only modified repository files, regenerating AST chunks and their embeddings without re-indexing the entire codebase. This synchronizes the Qdrant vector store efficiently as your code evolves.

Semantic search supports filtering results by chunk type, feature tags, and file patterns. After retrieving filtered vector matches, a re-ranking step refines the order to ensure the most relevant code snippets appear first.

The best way to orchestrate a vector store workflow is using a combined CLI and programmatic approach to manage provider configuration, Qdrant indexing, incremental embedding updates, and result formatting with re-ranking in one pipeline.