diffweave-ai 1.2.2

Assists with git workflows using LLMs.

DiffWeave AI

DiffWeave is a tool that automatically generates meaningful Git commit messages using large language models (LLMs). It analyzes your staged changes and creates descriptive commit messages, saving you time and ensuring consistent documentation.

Documentation available here

Installation & Quick Start

You can run DiffWeave via uvx diffweave-ai or from a local checkout using uv run diffweave-ai while developing this repo.

DiffWeave is installed as an isolated tool using uv:

# Make sure you have uv installed first
# https://docs.astral.sh/uv/getting-started/installation/

uvx diffweave-ai

Usage

Configure a model

Before generating commit messages, configure at least one model endpoint:

uvx diffweave-ai add-model \
  --model "name-of-your-model" \
  --endpoint "https://endpoint-url" \
  --token "$TOKEN"

Then set the default model to use:

uvx diffweave-ai set-model "name-of-your-model"

You can still override the model per invocation with the --model / -m flag.

Generate a commit message

Once you have a model configured and some changes staged in your current Git repository you can run:

uvx diffweave-ai

This will:

• Show the current git status.
• Optionally stage files for you (interactive by default).
• Generate a commit message using your configured model.
• Let you review/refine the message.
• Attempt git commit (and optionally git push and PR open if a URL is printed).

To specify a model for a single run:

uvx diffweave-ai -m "your-model-name"

If you prefer a simpler, more natural-language commit style rather than Conventional Commits, pass --simple:

uvx diffweave-ai --simple

You can also run in dry-run mode (generate and print a commit message without committing):

uvx diffweave-ai --dry-run

Or in a non-interactive mode, which will generate a message and attempt to commit and push without asking follow-up questions:

uvx diffweave-ai --non-interactive

Features

• AI-powered commit message generation based on staged diffs
• Interactive or non-interactive workflows
• Support for configuring custom LLM HTTP endpoints
• Ability to set and override the default model
• Optional simpler commit style via --simple
• Optional push and PR-open flow when git push prints a URL