Pullfrog is an open-source AI-driven GitHub bot created by Colin McDonnell, the author of Zod. It runs entirely inside GitHub Actions and doesn't depend on any hosted third-party service. If you need a self-controlled, model-agnostic tool for code review and automated CI fixes, Pullfrog is worth a look.

What Is Pullfrog

Simply put, Pullfrog is an AI Agent that runs inside your own GitHub Actions environment. It listens to repository webhook events (new PRs, Issues, CI failures, submitted reviews, and more) and automatically triggers the AI Agent to execute tasks.

Unlike hosted SaaS offerings such as CodeRabbit, Pullfrog's core characteristics are:

• Open source and free: The code is fully public and can be modified yourself
• Model-agnostic: Supports Anthropic, OpenAI, Google, Mistral, DeepSeek, OpenRouter, and more, switchable at any time
• Data never leaves your repository: All API keys are managed through GitHub Secrets, and the Agent executes in your repository's own Actions environment

Core Features

Pullfrog's capabilities cover more ground than ordinary code review tools:

• PR review: Automatically reviews pull requests and suggests improvements
• Issue triage: Automatically analyzes Issue content, categorizes, labels, and assigns owners
• Automated CI fixes: When a CI build fails, it automatically reads the logs, locates the problem, and submits a fix
• Merge conflict resolution: Automatically attempts to resolve conflicts when detected
• Plan generation: Automatically breaks large tasks into execution plans

It ships with a purpose-built MCP Server for executing git and GitHub operations (creating PRs, leaving reviews, reading CI logs, managing Issues). Shell commands run in isolated subprocesses with no access to sensitive environment variables. It also includes built-in headless browser tools, so the Agent can run end-to-end tests, take screenshots, and iterate on UI.

Quick Start

Installation

1. Install the Pullfrog GitHub App (apply on the Pullfrog website)
2. Add a pullfrog.yml workflow file to your repository

Usage

Two ways to trigger it:

• Manual trigger: @pullfrog in any Issue, PR, or comment to trigger an Agent run
• Automatic trigger: Configure automation trigger rules in the Pullfrog console (such as automatic review of new PRs, automatic fixes on CI failure, and so on)

Configuration Example

Create .github/workflows/pullfrog.yml in the repository root:

name: Pullfrog
on:
  issues:
    types: [opened, labeled]
  pull_request:
    types: [opened, synchronize]
  issue_comment:
    types: [created]
  pull_request_review:
    types: [submitted]

jobs:
  pullfrog:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Run Pullfrog
        uses: pullfrog/action@main
        with:
          api-key: ${{ secrets.PULLFROG_API_KEY }}
          llm-provider: anthropic  # can be switched to openai, google, etc.
          llm-model: claude-sonnet-4-20250514

Differences From Competitors

Use Cases

• Personal projects/small teams: Don't want to pay extra for code review; just use your existing LLM API key
• Code-security sensitive: Code and data never leave your GitHub environment
• Need more control: Want to customize review rules, choose models, and adjust Agent behavior

Caveats

• Pullfrog is currently in Beta, and features are still iterating rapidly
• It mainly targets CI/CD automation scenarios and is not suitable as a local development Agent (if you need a local Agent, just use Claude Code or OpenCode)
• The project's source code has earned more than 400 stars, and the community is fairly active

Project address: github.com/pullfrog/pullfrog