promptgitx 0.7.0

AI-powered Git commit assistant and PR analyzer.

PromptGitX

PromptGitX is an AI-powered Git commit assistant built as a Python CLI. It is intended to help generate clean commit messages, review staged changes, and manage LLM provider configuration from the terminal.

Features

• Interactive CLI built with Typer
• Rich terminal welcome screen
• LLM provider configuration for Groq, OpenAI, Anthropic, Gemini, and Ollama
• AI review reports for commits, commit ranges, pull requests, and staged changes
• LangGraph-based review workflow
• Hunk-based splitting for large file diffs instead of silent trimming
• Verified changed-line references from Git diff hunks
• Terminal, JSON, TXT, DOCX, and PDF report output

Project Structure

pyproject.toml          # PyPI/build metadata
src/
  promptgitx/
    main.py              # CLI entry point
    ai/                  # AI review/report helpers
    prompts/             # LangChain prompt templates
    config/              # LLM provider configuration helpers
    gitcodes/            # Git diff and repository helpers
    misc/                # Rich/figlet terminal UI helpers

Setup

1. Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate

2. Install dependencies:

pip install -r requirements.txt

For package-style local development, install the project in editable mode:

pip install -e .

3. Configure your LLM provider:

promptgitx config

How To Run

Recommended local development command after pip install -e .:

promptgitx

View available commands:

promptgitx --help

Show the installed version:

promptgitx --version

You can also run directly from source without installing:

python3 src/promptgitx/main.py

Or run it as a package module:

PYTHONPATH=src python3 -m promptgitx.main

Commands

Configure an LLM provider interactively:

promptgitx config

Configure a provider with CLI options:

promptgitx config --provider ollama --models llama3
promptgitx config --provider openai --models gpt-4o-mini --api-key YOUR_API_KEY

Reset configuration:

promptgitx config --reset

Start the chat command:

promptgitx chat

Generate a review report:

promptgitx analyze

Analyze examples:

promptgitx analyze --staged
promptgitx analyze --last
promptgitx analyze --last-n 3
promptgitx analyze --commit COMMIT_HASH
promptgitx analyze --compare main..feature-branch
promptgitx analyze --pr 123

Output modes:

promptgitx analyze --last --summary
promptgitx analyze --last --json
promptgitx analyze --last --save report.txt
promptgitx analyze --last --save report.json
promptgitx analyze --last --save report.docx
promptgitx analyze --last --save report.pdf

If --save is not passed, PromptGitX asks whether you want to save the report after displaying it. Use only one review target per command, such as --last or --staged, not both. --json prints raw JSON without the welcome banner so it can be used in scripts.

CLI Option Reference

Global options:

Long option	Short option	Description
--version	none	Show the installed PromptGitX version and exit.
--help	none	Show CLI help.

promptgitx config options:

Long option	Short option	Description
--provider	-p	LLM provider to configure: groq, openai, anthropic, gemini, or ollama.
--models	-m	Comma-separated list of up to five model names.
--api-key	none	API key for cloud providers.
--base-url	none	Base URL for local providers like Ollama.
--use	-u	Switch to an already configured provider.
--reset	-r	Reset saved configuration.

promptgitx analyze target options:

Long option	Short option	Description
--commit	-c	Review one specific commit.
--commits	-C	Review multiple commits. Can be provided more than once.
--compare	-p	Review a comparison such as main..feature-branch.
--pr	-P	Review a GitHub pull request by number. Requires GitHub CLI.
--last	-l	Review the latest commit.
--last-n	-n	Review the last N commits.
--staged	-s	Review staged changes.

promptgitx analyze output options:

Long option	Short option	Description
--json	none	Print the raw structured JSON report.
--summary	none	Print only the short review summary.
--save	none	Save the report to .txt, .json, .docx, or .pdf.

Configuration

PromptGitX supports the following providers:

• Groq
• OpenAI
• Anthropic
• Gemini
• Ollama

The config command writes provider settings to ~/.promptgitx/.env, including the current provider, API key or base URL, and up to five model names. This keeps PromptGitX configuration stable no matter which project directory you run the CLI from.

If you intentionally want a different config file, set PROMPTGITX_ENV_PATH to an explicit path before running PromptGitX.

The welcome screen displays the active model as:

Model: <PROVIDER> | <MODEL_NAME>

PromptGitX uses the first configured model for the active provider first. During one analyze run, if a model call fails, PromptGitX advances to the next configured model and keeps using that model for later chunks in the same run.

Analyze Workflow

The analyze command uses a LangGraph workflow:

load_diff
  -> parse_diff
  -> split_large_chunks
  -> review_chunks
  -> merge_file_reviews
  -> final_report

Workflow details:

• load_diff collects the requested Git diff using git or gh.
• parse_diff converts the raw diff into file-level chunks and computes changed-line references from diff hunk headers.
• split_large_chunks splits oversized file chunks into hunk-based review chunks without dropping diff data.
• review_chunks sends each review chunk to the configured LLM and validates returned references.
• merge_file_reviews combines subchunk reviews back into one review per file.
• final_report builds the structured report with risk, recommendation, grouped issues, and output-ready data.

Large diffs are split by hunks instead of being cut at an arbitrary character limit. If a single hunk is very large, PromptGitX keeps that hunk intact to avoid corrupting line references.

The final report is mostly built by deterministic Python logic. The LLM reviews individual chunks; Python then normalizes issues, validates references, merges file reviews, calculates risk, and formats the final outputs.

Build For Upload

Build source and wheel distributions:

python3 -m build --sdist --wheel

The files will be created in:

dist/

Status

This project is in early development. The review report workflow is functional, including saved TXT, JSON, DOCX, and PDF outputs. The chat command is still a placeholder.