Expand description
Repotoire - Graph-powered code analysis library
A fast, local-first code analysis tool that uses knowledge graphs to detect code smells, architectural issues, and technical debt.
Modules§
- ai
- AI-powered code fix generation
- baseline
- cache
- File caching for detectors
- calibrate
- Adaptive threshold calibration + predictive coding
- classifier
- Fast FP/TP classifier for findings
- cli
- CLI command definitions and handlers
- config
- Configuration module for Repotoire
- detectors
- Pure Rust code analysis detectors — zero external dependencies.
- engine
- Analysis engine — layered, presentation-free code health analysis.
- fixes
- Rule-based fix suggestions
- git
- Git history analysis module
- graph
- Graph database for code analysis.
- http
- Minimal HTTP/1.1 client over TcpStream + rustls. Replaces ureq (~41 transitive deps) for 4 call sites. Supports both http:// (Ollama localhost) and https:// (cloud APIs).
- log
- Minimal tracing subscriber that writes to stderr with env filter support.
- models
- Core data models for Repotoire
- parsers
- Source code parsers using tree-sitter
- predictive
- Hierarchical Predictive Coding Engine
- quantize
- TurboQuant vector quantization (Zandieh et al. 2025).
- reporters
- Output reporters for Repotoire analysis results
- scoring
- Graph-Aware Health Scoring System
- telemetry
- tui
- Minimal TUI framework for the findings browser. Replaces ratatui + crossterm (~100 transitive deps) with ~400 lines + libc.
- values
- Graph-based constant propagation — static value oracle for detectors.