Core Concepts

To get the most out of CodeBroker, it helps to understand how it models your codebase under the hood.

Repository Indexing

CodeBroker scans your repository and parses the syntax trees of all supported source files using Tree-sitter. It extracts functions, classes, interfaces, and variables.

Semantic Bindings

Unlike simple AST parsers, CodeBroker resolves imports and exports to create a unified semantic graph across your entire codebase.

PageRank & Communities

By applying graph algorithms like PageRank and Louvain community detection, CodeBroker identifies architectural boundaries and critical files.

Context Capsules

When an AI needs context, CodeBroker calculates the shortest paths between relevant symbols and returns a perfectly trimmed, highly relevant markdown payload.

Embeddings

While the semantic graph is deterministic, CodeBroker also generates vector embeddings for every node in the graph. This hybrid approach means you can search by exact symbol name (deterministic) or by natural language concept (probabilistic), and CodeBroker will use the graph edges to fetch the related implementations.

Hybrid Retrieval
When querying, use natural language. The engine automatically maps your intent to the correct semantic nodes and traverses the graph.