codedupes

codedupes detects duplicate and potentially unused Python code with:

• Traditional AST/token matching (exact + Jaccard near-duplicate)
• Semantic matching with model-profile embeddings (default gte-modernbert-base)
• Heuristic unused-code detection

Install

pip install "codedupes @ git+https://github.com/pszemraj/codedupes.git"

Optional GPU extras:

pip install "codedupes[gpu] @ git+https://github.com/pszemraj/codedupes.git"

Requires Python 3.11+. Details are in
docs/install.md

Quick Start

codedupes check ./src
codedupes search ./src "normalize request payload"
codedupes info

codedupes check defaults to a hybrid-first report:

• one combined duplicate list (Hybrid Duplicates)
• likely dead code (potentially_unused)

Use --show-all to include raw traditional + raw semantic duplicate lists.

Documentation

Primary docs live under docs/:

• docs/index.md: documentation map and ownership
• docs/cli.md: commands, flags, and defaults
• docs/model-profiles.md: semantic model aliases, profile defaults, and task behavior
• docs/analysis-defaults.md: analysis-behavior defaults and heuristics
• docs/output.md: JSON schemas and exit codes
• docs/usage.md: practical workflows and tuning examples
• docs/python-api.md: programmatic API usage
• docs/hybrid-tuning.md: hybrid gate tuning workflow

Notes and limits

• Call graph and unused detection are heuristic and conservative by default.
• Semantic model-profile defaults and task behavior are defined in
  docs/model-profiles.md.
• Analysis defaults (semantic candidate scope, tiny-traditional filtering, hybrid gates) are defined in
  docs/analysis-defaults.md.
• Semantic analysis may download model weights on first use.
• Extraction skips common artifact/cache directories by default (__pycache__, .venv, etc).