Codet

A cross-platform command-line tool for Git repository analysis. Profile commit history, detect code hotspots, generate diff reports, and optionally leverage AI for deeper insights. The generated git diff files integrate seamlessly with Cursor for collaborative development.

Features

• Commit history analysis - browse and filter recent commits by time range, author, email, keyword, or commit hash
• Code hotspot detection - identify frequently modified files and directories to spot active areas
• Flexible search modes - union (match any condition) or intersection (match all conditions)
• Diff report generation - produce aggregated .diff report files for review in Cursor or other tools
• AI-powered analysis - optional OpenAI / Azure OpenAI integration for automated commit summarization
• Interactive dashboard - Plotly Dash visualization via the codet-dash command
• JSON export - structured per-commit JSON output for downstream pipelines (-oj)
• Cross-platform - works on Windows, macOS, and Linux

Installation

From PyPI

pip install codet

From source

git clone https://github.com/host452b/codet.git
cd codet
pip install -e .

Development dependencies

pip install -e ".[dev]"

Quick Start

# analyze the current repo, last 7 days, with hotspot analysis
codet -d 7 -s

# search for keywords "Triton" and "cuda" in a cloned repo
git clone https://github.com/pytorch/pytorch.git
codet -d 7 -k Triton -k cuda -s -p pytorch

# filter by author email
codet -d 14 -e alice@example.com

# filter by commit hash (supports partial match)
codet -c abc1234

# intersection mode: must match ALL conditions
codet -d 7 -k feature -e dev@example.com -m intersection

CLI Reference

usage: codet [-h] [--version] [-d DAYS] [-e EMAIL] [-u USER] [-k KEYWORD]
             [-c COMMIT] [-g] [-r] [-p PATH] [-s] [-m {union,intersection}]
             [-mo MODEL] [-to API_TOKEN] [-oe OPENAI_ENDPOINT]
             [-cp CUSTOM_PROMPT] [-f INPUT_FILE] [-oj]

Basic Options

Flag	Description	Default
-d, --days	Look back N days for commits	30
-e, --email	Filter by author email (repeatable)	—
-u, --user	Filter by author name (repeatable)	—
-k, --keyword	Search keyword in commit diffs (repeatable)	—
-c, --commit	Filter by commit hash, supports partial match (repeatable)	—
-g, --debug	Enable debug logging	False
-r, --recursive	Recursively scan subdirectories for git repos	True
-p, --path	Path to analyze	current directory
-s, --hotspot	Enable code hotspot analysis	True
-m, --mode	Search mode: union or intersection	union

AI Options

Codet works with any OpenAI-compatible API — just pass the endpoint and model name:

Flag	Description	Default
-mo, --model	Model name as required by your provider (e.g., gpt-4.1, openai/openai/gpt-5-nano, llama3)	—
-to, --api-token	Bearer token (or set AI_API_TOKEN env var). If not set, AI analysis is skipped	—
-oe, --openai-endpoint	Base URL of the API (/v1 is appended automatically if missing)	—
-cp, --custom-prompt	Custom prompt appended to AI analysis	—
-f, --input-file	Additional file to include in AI analysis	—
-oj, --output-cook-json	Export per-commit JSON reports to json_cook/	False

Examples with different providers:

# Standard OpenAI
codet -d 10 -s -to $OPENAI_KEY -oe "https://api.openai.com/v1" -mo "gpt-4.1"

# Local Ollama
codet -d 10 -s -to ollama -oe "http://localhost:11434/v1" -mo "llama3"

Dashboard

Codet includes an interactive Plotly Dash dashboard for visualizing analysis results.

# launch dashboard with JSON data directory
codet-dash --json-path json_cook/

Output

Diff Report

Each run generates a git_patch_report_<timestamp>.diff file containing aggregated patches with contextual metadata. Open it directly in Cursor for AI-assisted code review.

JSON Export

With -oj, per-commit JSON files are written to json_cook/<repo_name>/, each containing commit metadata, changed files, and optional AI summaries.

Development

git clone https://github.com/host452b/codet.git
cd codet
pip install -e ".[dev]"

# lint
flake8 . --max-line-length=127

# format
black .
isort .

# test
pytest

License

MIT License. See LICENSE for details.