Skip to content

Create and apply diffs with AI

Modify projects with natural language

gptdiff "Add button animations on press" --apply

✅ Successfully applied patch
gptdiff "Add API documentation" --call

🔧 Patch written to diff.patch
gptdiff "Improve error messages"

📄 LLM not called, written to prompt.txt
gptdiff "Improve login error messages" auth.py --apply

✅ Patch applied: auth.py updated successfully

Set up agents loops with simple bash commands

while
do
  gptdiff "Add missing test cases" --apply
done

open source and MIT licensed at github.com/255BITS/gptdiff

🚀 Make impactful changes with one command

  • --apply AI-powered patch recovery
  • --nobeep Disable completion notifications
  • --temperature Control creativity (0-2)
  • --model Switch between LLM providers

Features

  • Describe changes in plain English
  • AI gets your whole project
  • Auto-fixes conflicts
  • Keeps code functional
  • Fast setup, no fuss
  • You approve every change
  • Costs are upfront

Ready to simplify your workflow? Try GPTDiff now!

How to develop with gptdiff

🚀 The new dev cycle

  1. Generate Changes

    gptdiff "Add type safety" --apply

  2. Review Modifications

    git add -p

  3. Finalize Changes

    git commit -m 'AI improvements'

  4. Stash Uncommitted Changes

    git stash

Getting Started

For a step-by-step guide on how to install and configure gptdiff, check out the Installation Guide.

$ pip install gptdiff
🔧 Built with AI Agent Toolbox

Using gptpatch

You can further refine your diff application using gptpatch. For example, to enhance clarity:

gptdiff "make the value prop clearer" splash.html

Then, copy the generated prompt from prompt.txt to your clipboard and send it to OpenAI. Once you receive the outputs, open patch.diff, paste the outputs, and finally run:

gptpatch patch.diff

Note: You can configure GPTDIFF_SMARTAPPLY_ variables independently from the base GPTDIFF_ variables.

Using Anthropic Claude with Extended Thinking

When working with Anthropic Claude models, you can enable the extended thinking capability using the --anthropic_budget_tokens flag: