maestro

Maestro is a production-grade RAG engine designed to act as an intelligent gateway between software agents and a massive library of specialized technical knowledge. By indexing hundreds of domain-specific skills—ranging from Swift, SwiftUI, and concurrency patterns to App Store Connect workflows and architecture—Maestro prevents context window degradation. Instead of loading every possible skill into the LLM context, it executes a high-speed vector search to retrieve only the relevant, actionable chunks required for the specific task at hand.

• Employs a hybrid RAG pipeline using ChromaDB, BM25 lexical search, and neural embeddings for high-precision retrieval.

• Features a concept graph to expand search queries (e.g., mapping "data race" to "actor isolation" and "thread safety").

• Implements iterative diffusion reranking and cross-encoder scoring to ensure retrieved knowledge is mathematically relevant to the query.

• Supports seamless integration with Claude Code via the Model Context Protocol (MCP).

• Provides a complete CLI toolchain for indexing, status reporting, and debugging search pipeline performance.

• Intended for developers and autonomous agents working on complex iOS, macOS, and multi-platform Swift projects.

• Typical usage involves calling search_skills() before any coding task to retrieve precise architectural guidance or framework-specific patterns.

• Can operate in offline/manual modes if MCP is unavailable, allowing developers to generate and paste context blobs directly into an LLM.

• Optimized for performance, consistently returning high-relevance chunks in under 100ms.

• Constraints include requiring a Python 3.11+ environment and the initial setup of skill directories for indexing.