An AI-powered runtime debugging agent that embeds directly into your Ruby application. Add one gem, configure an LLM key, and chat with your live app at /agent to inspect GC, ObjectSpace, threads, routes, Redis, Rails models/routes, Sidekiq queues, Puma stats, fibers/signals, process info, HTTP requests, and more — 98 diagnostic tools across 36 inspectors.

Requires Ruby 2.7+ for pattern matching guards. Framework inspectors (Rails, Sidekiq, Puma) are optional and auto-detected via defined?.

# Gemfile
gem 'debug-agent', github: 'topcheer/ruby-debug-agent'

Or install directly:

gem install debug-agent --source https://github.com/topcheer/ruby-debug-agent

require 'sinatra/base'
require 'debug_agent'

class MyApp < Sinatra::Base
  # One line to integrate
  register DebugAgent::Middleware
end

export LLM_API_KEY=your-key
export LLM_BASE_URL=https://open.bigmodel.cn/api/coding/paas/v4  # default
export LLM_MODEL=glm-5.2                                          # default

Supports any OpenAI-compatible endpoint.

http://localhost:4567/agent

• Streaming AI responses with real-time tool call badges (pending / success / error)
• Context compression — automatically summarizes old conversation when token limit is approached
• Dark-themed chat UI with full markdown rendering (tables, code blocks, lists)
• Max tool rounds (25) with forced final summary when limit is reached
• 98 diagnostic tools across 36 inspectors
• Zero external dependencies (no Datadog, no Grafana, no APM)

require 'debug_agent'

DebugAgent.register_tool('check_redis', 'Check Redis connection') do
  { connected: true }
end

The demo uses Sinatra + redis-rb + SQLite + Sidekiq. Start Redis with Docker Compose first:

# docker-compose.yml
services:
  redis:
    image: redis:7-alpine
    ports:
      - "6379:6379"
    command: redis-server --save 60 1 --loglevel warning

docker compose up -d

export LLM_API_KEY=your-key
cd demo && ruby -I../lib app.rb
# Open http://localhost:4567/agent

This project was built using ggcode — an AI coding assistant for terminal-based development.

MIT