KiroGraph: Local code knowledge graph for AI, optimized for token efficiency (github.com)

Kiro has announced the launch of KiroGraph, a local semantic code knowledge graph designed to optimize token efficiency for AI-powered coding tasks. This new feature allows Kiro to instantly access symbol relationships, call graphs, and type hierarchies by querying a pre-indexed graph stored in a local SQLite database, significantly reducing the number of tool calls and speeding up responses for complex queries. Unlike traditional methods that rely on scanning files and performing multiple tool calls, KiroGraph leverages tree-sitter to generate an Abstract Syntax Tree (AST) from source files, streamlining the process of understanding the codebase. The introduction of KiroGraph is noteworthy for the AI/ML community as it enhances the capability of coding assistants by maintaining complete data privacy—no information is sent externally. The architecture of KiroGraph supports three layers of indexing (structural, semantic, and architecture), integrating seamlessly with Kiro's existing tools for natural language and semantic searches. By producing rich 768-dimensional vector embeddings for symbols and allowing various embedding engines for optimized searching, KiroGraph marks a significant step forward in the integration of AI and machine learning within code development environments, promising to empower developers with more efficient workflows and insights into their codebases.