Ralphex – autonomous GPT Codex agent loop for ChatGPT Pro users
Ralph fork for ChatGPT Pro instead of Claude, but requires prd.json boilerplate.

Agent self-bootstraps with no frameworks, but novelty wears fast without real problem solved.
Developers curious about agent mechanics, agentic LLM loop fundamentals, or OS/infrastructure builders.
LangChain tutorials · AutoGen minimal examples · Agent loop playground projects
The result is a model-specific REPL "genesis snippet." It's a short bash script that prompts for your API key, asks the model to generate a REPL with an agentic tool-use loop, and then drops you straight into that running REPL.
The self-imposed rules:
1. No copy-pasting code after the initial snippet. 2. No manual editing—the agent has to write and fix everything for you. 3. Zero agent frameworks or libraries.
From there, the goal is to ask the agent to rewrite its own REPL, add quality-of-life improvements, and eventually figure out how to connect itself to Telegram.I put the snippets up on a site along with some open challenges like code golf and speed runs:
To try it, you just need your own API key and a Linux VM or Docker container. If you want to see how it actually behaves before running it, here is the raw terminal output of my first successful run:
Ralph fork for ChatGPT Pro instead of Claude, but requires prd.json boilerplate.
Founding team of AI agents, but existing agentic platforms already do this orchestration.
A polished directory of aesthetic Unicode combos, but it's just a static list.
Compelling airplane-mode story, but the 'guide' is unverified claims dressed as a product.
Git worktree wrapper for AI agents, but worktree management is niche and solved adequately by git natively.
Narrative-first AI content beats technical novelty; feels like marketing for AgentHost.