Semantic Search & Local Context
AI that understands your whole codebase. Without uploading a single file.
Most cloud-based tools require uploading your entire codebase to external servers for indexing. Reflexor's semantic RAG engine chunks files, computes vector embeddings, and answers complex multi-file architectural queries completely on your local machine.
How Local Semantic Retrieval Works
- Intelligent AST-Aware Chunking: Code files are split along natural function, class, and interface boundaries to preserve semantic cohesion.
- Accelerated Local Embeddings: Compact, high-density neural embedding models run with hardware acceleration for near-instant index builds.
- Hybrid Retrieval (Dense + Sparse): Merges semantic vector similarity with exact code identifier matching for surgical precision in responses.
Technical Highlights
- Absolute Codebase Privacy: Not a single file, code snippet, or vector embedding is ever communicated over the public internet.
- Incremental Change Tracking: Only modified files are re-embedded on save, ensuring near-zero background CPU and battery drain.
Frequently Asked Questions about Local RAG
How much disk space does the local RAG index consume?
The vector index typically uses only 2% to 5% of the total codebase size and is stored inside the extension's local application cache directory.
Can Reflexor answer questions about files I don't have open?
Yes! The RAG system searches across all indexed workspace files, enabling the AI to pull context from shared libraries, utility functions, and distant modules.
Are large monorepos supported?
Yes. Reflexor strictly honors `.gitignore` rules and supports inclusion/exclusion patterns to index only your active project folders efficiently.