pullfrog/pullfrog

Open-source model-agnostic BYOK GitHub bot that runs in GitHub Actions

What it solves

Pullfrog is a GitHub Actions-based automation tool that brings AI coding agents into the software development lifecycle. It eliminates the need for a separate, managed AI service by allowing developers to run agents directly within their own CI/CD pipeline, using their own API keys or existing subscriptions (BYOK - Bring Your Own Key).

How it works

Pullfrog acts as a wrapper for existing coding agents like Claude Code, Codex, and OpenCode. It listens for GitHub events (such as PRs opened, issues created, or CI failures) and triggers agent runs based on a pullfrog.yml configuration. It provides the agent with a purpose-built MCP server for GitHub operations, a secure isolated shell, and a headless browser for UI testing. All operations use short-lived, auto-revoking GitHub tokens to maintain security.

Who it’s for

Software engineers and DevOps teams who want to automate code reviews, issue triage, and bug fixing within GitHub, while maintaining full control over their infrastructure, keys, and model costs.

Highlights

  • BYOK Flexibility: Supports various LLM providers (Anthropic, OpenAI, Google, xAI, etc.) and existing subscriptions like Claude Pro/Max.
  • Agentic Automations: Automatically reviews PRs, addresses review comments, fixes CI failures, and resolves merge conflicts.
  • GitHub Native: Integrates as a GitHub Action, using native permissions and a pullfrog.yml workflow.
  • Developer Tools: Includes a headless browser for screenshots and UI iteration, and support for structured JSON output via JSON Schema.
  • Ad hoc Prompting: Allows tagging @pullfrog in issues or comments to trigger immediate agent actions.

Related

  • Project
    hustcer/deepseek-review

    An automated code review tool that uses DeepSeek AI to analyze GitHub Pull Requests and local Git changes for bugs and security issues.

  • Project
    anthropics/claude-code-base-action

    A GitHub Action that integrates Claude Code into CI/CD workflows, enabling automated agentic coding tasks like PR reviews and dependency updates.

  • Project
    patched-codes/patchwork

    A self-hosted CLI agent that automates development tasks like PR reviews, bug fixing, and security patching using customizable LLM-assisted workflows.

  • Project
    github/accessibility-scanner

    The AI‑powered Accessibility Scanner is a GitHub Action that automatically checks web pages for WCAG issues, creates GitHub issues for each finding, and can use GitHub Copilot to draft fix pull‑requests. It integrates into any repo’s Actions workflow, supports custom scans, screenshots, and grouping, but is still in public preview and AI fixes require human review.

  • Project
    taovc/pr-cockpit

    A local AI-powered PR review and management tool that uses isolated git worktrees to analyze, fix, and manage GitHub pull requests using Claude or Codex.