sia-code

To search an unfamiliar codebase for architecture analysis, you can use local-first codebase intelligence to perform lexical-first BM25 search and multi-hop research. This traces code relationships to map dependencies and identify implementation patterns across 12 supported languages.

Tracing authentication flows across a complex project is best handled by multi-hop research, which traverses code relationships to answer complex questions. This approach identifies specific implementation patterns and error handling mechanisms without requiring prior knowledge of the codebase.

Local-first code search supports AST parsing across 12 programming languages. This enables accurate code analysis and dependency mapping, while offering optional semantic search via embeddings to supplement the lexical-first BM25 and FTS5 search mechanisms.

To initialize and index a local project for code research, run `uvx sia-code init` followed by `uvx sia-code index .` in your project directory. This builds the local index required for lexical search and multi-hop research.

You can store and retrieve historical technical decisions using built-in project memory. This feature records changelogs and architectural insights locally, allowing you to maintain context and track historical decisions while researching complex or unfamiliar codebases.

Lexical search should be used as the primary method for code analysis because it utilizes BM25 and FTS5 to achieve 89.9% Recall@5. Semantic search via embeddings is optional and supplements lexical search when deeper contextual matching is required.