itsharex / pr-agent Goto Github PK

Making pull requests less painful with an AI agent

PR-Agent offers extensive pull request functionalities across various git providers:

Review the usage guide section for detailed instructions how to use the different tools, select the relevant git provider (GitHub, Gitlab, Bitbucket,...), and adjust the configuration file to your needs.

You can try GPT-4 powered PR-Agent, on your public GitHub repository, instantly. Just mention @CodiumAI-Agent and add the desired command in any PR comment. The agent will generate a response based on your command. For example, add a comment to any pull request with the following text:

@CodiumAI-Agent /review

and the agent will respond with a review of your PR

To set up your own PR-Agent, see the Installation section below.

To get started with PR-Agent quickly, you first need to acquire two tokens:

1. An OpenAI key from here, with access to GPT-4.
2. A GitHub personal access token (classic) with the repo scope.

There are several ways to use PR-Agent:

• Method 1: Use Docker image (no installation required)
• Method 2: Run from source
• Method 3: Run as a GitHub Action
• Method 4: Run as a polling server
  • Request reviews by tagging your GitHub user on a PR
• Method 5: Run as a GitHub App
  • Allowing you to automate the review process on your private or public repositories
• Method 6: Deploy as a Lambda Function
• Method 7: AWS CodeCommit

The following diagram illustrates PR-Agent tools and their flow:

Check out the PR Compression strategy page for more details on how we convert a code diff to a manageable LLM prompt

A reasonable question that can be asked is: "Why use PR-Agent? What make it stand out from existing tools?"

Here are some advantages of PR-Agent:

• We emphasize real-life practical usage. Each tool (review, improve, ask, ...) has a single GPT-4 call, no more. We feel that this is critical for realistic team usage - obtaining an answer quickly (~30 seconds) and affordably.
• Our PR Compression strategy is a core ability that enables to effectively tackle both short and long PRs.
• Our JSON prompting strategy enables to have modular, customizable tools. For example, the '/review' tool categories can be controlled via the configuration file. Adding additional categories is easy and accessible.
• We support multiple git providers (GitHub, Gitlab, Bitbucket, CodeCommit), multiple ways to use the tool (CLI, GitHub Action, GitHub App, Docker, ...), and multiple models (GPT-4, GPT-3.5, Anthropic, Cohere, Llama2).
• We are open-source, and welcome contributions from the community.

• Support additional models, as a replacement for OpenAI (see here)
• Develop additional logic for handling large PRs (see here)
• Add additional context to the prompt. For example, repo (or relevant files) summarization, with tools such a ctags
• PR-Agent for issues, and just for pull requests
• Adding more tools. Possible directions:
  • PR description
  • Inline code suggestions
  • Reflect and review
  • Rank the PR (see here)
  • Enforcing CONTRIBUTING.md guidelines
  • Performance (are there any performance issues)
  • Documentation (is the PR properly documented)
  • ...

• CodiumAI - Meaningful tests for busy devs (although various capabilities are much more advanced in the CodiumAI IDE plugins)
• Aider - GPT powered coding in your terminal
• openai-pr-reviewer
• CodeReview BOT
• AI-Maintainer