AI Powered Commit Message Normalization Tool
Project description
CMAI - AI-Powered Commit Message Normalizer
CMAI is a CLI tool that uses AI to transform informal commit descriptions into standardized, professional Git commit messages based on your staged changes.
🚀 Quick Start
1. Installation
Install via uv (recommended) or pip:
# Install with all providers support
uv tool install cmai[all-providers]
# OR
pip install cmai[all-providers]
# For specific providers only: cmai[openai], cmai[ollama], cmai[anthropic], etc.
2. Configuration
Create the configuration file at ~/.config/cmai/settings.env:
# --- Remote Provider Example (OpenAI, DeepSeek, Zai, etc.) ---
PROVIDER=openai
API_KEY=your_api_key_here
MODEL=gpt-4o-mini
# API_BASE=... (Optional: Only needed for non-standard endpoints)
# --- Anthropic (Claude) Provider Example ---
# PROVIDER=anthropic
# API_KEY=your_anthropic_api_key_here
# MODEL=claude-3-5-sonnet-20241022
# ENABLE_THINKING=true
# THINKING_BUDGET=1024
# --- Local Provider Example (Ollama) ---
# PROVIDER=ollama
# OLLAMA_HOST=http://localhost:11434
# MODEL=qwen2.5:7b
# --- Commit Specification Rules ---
COMMIT_SPEC=conventional
COMMIT_STRICT=true
# COMMIT_ALLOWED_TYPES=feat,fix,docs,chore
COMMIT_SCOPE_POLICY=optional
COMMIT_SUBJECT_MAX_LEN=72
COMMIT_HEADER_MAX_LEN=100
COMMIT_SUBJECT_CASE=lower
COMMIT_ALLOW_BANG=true
# --- Large Diff and Context Optimization ---
MAX_DIFF_LENGTH=8000
MAX_DIFF_FILE_LINES=50
MAX_DIFF_FILES_FOR_AI=30
ENABLE_DIFF_LOCAL_SUMMARY=true
ENABLE_SPLIT_SUGGESTION=true
SPLIT_CONFIDENCE_THRESHOLD=0.75
DIFF_SUMMARY_CONCURRENCY=4
RETRY_MAX_ATTEMPTS=3
RETRY_BASE_DELAY_SECONDS=1.0
RETRY_MAX_DELAY_SECONDS=8.0
Supported Providers: openai, bailian, deepseek, siliconflow, anthropic, claude, zai (智谱), ollama.
Tip: You can also set CMAI_API_KEY or ANTHROPIC_API_KEY as environment variables instead of putting them in the config file.
3. Usage
Stage your changes and run cmai:
git add .
cmai "fix a bug"
The tool will output a normalized message and prompt for action:
- [c]ommit: Execute git commit.
- [e]dit: Edit the message manually.
- [r]egenerate: Ask the model to regenerate. You can provide an optional additional prompt.
- [a]bort: Cancel.
When COMMIT_STRICT=true, non-compliant messages cannot be committed.
The CLI will show warnings and only allow edit, regenerate, or abort until the message passes validation.
For very large staged diffs, CMAI now falls back to per-file truncated diff previews instead of only file names. When enabled, CMAI also performs file-level AI summaries and can suggest splitting unrelated staged changes into separate commits. During file summarization, progress lines now display only the file name (not full path).
When providers hit rate limits (for example 403 RPM limit exceeded or 429), CMAI automatically retries with exponential backoff.
If final commit generation still fails after retries, CMAI falls back to a local heuristic commit message instead of exiting immediately.
If a commit error occurs, an appropriate error message will be displayed and retained here. You can open a new terminal to fix these issues and then input 'c' to proceed with the commit.
🛠 CLI Options
cmai [MESSAGE] [OPTIONS]
Options:
-c, --config TEXT Path to a custom configuration file
-r, --repo TEXT Path to the git repository (default: current dir)
-l, --language TEXT Target language for the commit message (e.g., "Chinese")
✅ Commit Specs and Formatting Preferences
COMMIT_SPEC:conventionalorangularCOMMIT_STRICT: iftrue, block commit until message is validCOMMIT_ALLOWED_TYPES: optional comma-separated override for allowed typesCOMMIT_SCOPE_POLICY:optional,required, orforbidCOMMIT_SUBJECT_MAX_LEN: max subject lengthCOMMIT_HEADER_MAX_LEN: max full header lengthCOMMIT_SUBJECT_CASE:lower,sentence, oranyCOMMIT_ALLOW_BANG: whether!is allowed in headerMAX_DIFF_LENGTH: max characters for raw staged diff contextMAX_DIFF_FILE_LINES: per-file changed lines kept in truncated preview modeMAX_DIFF_FILES_FOR_AI: max files included in file-level AI summarizationENABLE_DIFF_LOCAL_SUMMARY: enable two-stage AI diff summarizationENABLE_SPLIT_SUGGESTION: enable split-commit recommendationSPLIT_CONFIDENCE_THRESHOLD: minimum AI confidence to show split recommendationDIFF_SUMMARY_CONCURRENCY: concurrent file-summary requestsRETRY_MAX_ATTEMPTS: max attempts when provider hits rate limitRETRY_BASE_DELAY_SECONDS: initial backoff delay for rate-limit retryRETRY_MAX_DELAY_SECONDS: max backoff delay for rate-limit retry
🔁 Retry and Fallback Behavior
- CMAI retries only on likely rate-limit errors (such as
429,RPM limit,too many requests,limit exceeded). - Backoff uses exponential delays:
base * 2^(attempt-1), capped byRETRY_MAX_DELAY_SECONDS. - If retries are exhausted for final commit generation, CMAI builds a local commit message that still follows your configured commit rules.
📦 Development
git clone [https://github.com/yumuzhihan/cmai.git](https://github.com/yumuzhihan/cmai.git)
cd cmai
uv sync --all-extras # Or other groups
python -m pytest # Run tests
📄 License
This project is licensed under the MIT License.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file cmai-0.2.5.tar.gz.
File metadata
- Download URL: cmai-0.2.5.tar.gz
- Upload date:
- Size: 32.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
c3a0e3409c8ba310ce23441dd8a2af8ac35c2133fd460db0c25df0a9ffd0972b
|
|
| MD5 |
b21164b472bebad2a59ccf22d8a3c7f7
|
|
| BLAKE2b-256 |
a06b150aad6f188e31303c608b2aaf05933ef767f734250c8daea7b97d29659a
|
File details
Details for the file cmai-0.2.5-py3-none-any.whl.
File metadata
- Download URL: cmai-0.2.5-py3-none-any.whl
- Upload date:
- Size: 33.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
14efd45ae62e6cf3fc6d2d29a46f05c54da0dedd0e0b5d3b1692886d01c06798
|
|
| MD5 |
11c2d1a495728ecf59f7e74388b9fd35
|
|
| BLAKE2b-256 |
bba89735a060ef04eea7f73885752558d6396f7e199e126a94ae846f27deb3ef
|