prgen

Generate a pull request title and description from git diff and commit history.

Author: Jean Paul Fernandez · github.com/jpxoi/prgen

Licensed under the GNU General Public License v3.0 (GPL-3.0-only).

Requirements

• Python 3.10+
• git on your PATH
• One of:
  • OpenAI API access
  • Google Gemini API access
  • A local or remote Ollama server

Install

From PyPI:

pip install prgen-cli
# or
uv tool install prgen-cli

From a clone:

uv sync

Run the CLI with:

prgen --help

What It Does

prgen compares HEAD against a base ref, collects:

• git diff <base>...HEAD
• git log <base>..HEAD

It sends that context to an LLM and prints:

• a PR title from <summary>...</summary>
• a PR description from <body>...</body>

If the model does not return those tags, prgen prints the raw model output instead.

Providers

prgen supports three backends:

• openai
• gemini
• ollama

--provider auto is the default.

In auto mode:

• Gemini is chosen when GOOGLE_API_KEY is available
• otherwise OpenAI is chosen when OPENAI_API_KEY is available
• if neither key is configured, prgen exits with an error

Ollama is always explicit:

• use --provider ollama
• also pass --model <name>
• --tier presets do not apply to Ollama

Quick Start

OpenAI:

prgen config set OPENAI_API_KEY sk-...
prgen

Gemini:

prgen config set GOOGLE_API_KEY your-key
prgen --provider gemini

Ollama:

prgen --provider ollama --model llama3.1:8b

If the Ollama model is missing locally, let prgen pull it:

prgen --provider ollama --model llama3.1:8b --pull

Configuration

Configuration lives in ~/.config/prgen/config.json.

If XDG_CONFIG_HOME is set, prgen uses:

$XDG_CONFIG_HOME/prgen/config.json

You can manage the file with:

prgen config
prgen config show
prgen config path

Supported persisted keys:

• OPENAI_API_KEY
• GOOGLE_API_KEY
• OLLAMA_HOST
• base
• provider
• tier

Notes:

• OPENAI_API_KEY and GOOGLE_API_KEY are treated as secrets
• OLLAMA_HOST is not secret and is merged into the environment if set
• base, provider, and tier are optional CLI defaults
• model and context are not persisted in config

Examples:

prgen config
prgen config set OPENAI_API_KEY sk-...
prgen config set GOOGLE_API_KEY your-key
prgen config set OLLAMA_HOST http://127.0.0.1:11434
prgen config set base origin/main
prgen config set provider ollama
prgen config set tier pro
prgen config unset OLLAMA_HOST
prgen config show

To read a secret value from stdin:

prgen config set OPENAI_API_KEY - < key.txt

Defaults

Built-in defaults:

Config-file defaults apply only when you omit the matching flag:

• base
• provider
• tier

Explicit flags always win over the config file.

Current model presets:

• OpenAI default: gpt-5-mini
• OpenAI pro: gpt-5.4
• Gemini default: gemini-3-flash-preview
• Gemini pro: gemini-3.1-pro-preview

Usage

Basic usage:

prgen

Pick a different base:

prgen --base main

Run against another repository:

prgen -C ~/src/my-project

Override the model directly:

prgen --provider openai --model gpt-5.4
prgen --provider gemini --model gemini-3.1-pro-preview
prgen --provider ollama --model mistral-small3.1

Add extra context:

prgen --context "Focus on customer-facing impact and rollout notes."

Inspect API response headers (useful on rate-limit errors):

prgen --provider openai --verbose

With --verbose, prgen prints to stderr:

• provider, model, git base, commit count, and prompt size
• token usage (OpenAI / Gemini) or eval counts (Ollama)
• HTTP status and useful response headers (rate limits, request id)

Transport headers are omitted. Authorization values and API keys in URLs are redacted.

Behavior Notes

• prgen validates that --base resolves before generating anything
• if there are no commits and no file changes vs the base ref, prgen exits with an error
• when --provider ollama --pull is used, prgen can download the model automatically
• when stderr is a TTY, loading states and Ollama downloads use Rich UI output
• --verbose prints usage and API response headers after the call (or after an error), including rate-limit metadata when the provider sends it

Development

Install local dependencies:

uv sync

Format and lint:

uv run ruff format .
uv run ruff check .

Run from the repo without installing globally:

uv run prgen --help

Install the local checkout as a global tool:

uv tool install .
# or
pipx install .