What it does

Diffron writes Conventional Commits style messages from staged changes and drafts pull request titles and descriptions from branch diffs. It uses a local Lemonade LLM server, which the page describes as AMD's local server for Ryzen AI PCs, so no cloud service is required. Setup means installing the Lemonade server, then lemonade-sdk and Diffron with pip, optionally choosing a model with diffron-setup-model, and installing Git hooks so messages are generated at commit time. It also works through GitHub Desktop 3.5.5 or newer, and diffron status reports the state of Lemonade, the hooks and AI-agent detection. Diffron skips generation when it detects an AI coding agent committing, by checking environment variables, Git config and whether the message already follows Conventional Commits. It runs on Windows, Linux and macOS, has a Python API, and is MIT licensed.

Capabilities

Conventional Commits messages from staged changes
PR titles and descriptions from branch diffs
Uses a local Lemonade LLM server, no cloud required
Git hooks and GitHub Desktop support
Skips commits made by AI coding agents
Curated model list with a setup command

When to use it

You want commit messages generated without sending diffs to a cloud service
You already run, or plan to run, a Lemonade local server
You want a PR description drafted from a branch diff