Stop stressing over not having an agent running. Ralph is always running
Put your AI coding agent in a while True loop.
Write a prompt. Walk away. Wake up to commits.
uv tool install ralphify
The scene
Monday morning standup.
Everyone's talking about what their agent shipped over the weekend.
“My agent landed 4 PRs while I was at brunch.”
“I left Codex running overnight. Woke up to a clean backlog.”
“I... opened my laptop just now.”
Agent FOMO is real. But it doesn't have to be.
The fix
One command.
Weekend shipped.
ralph runHow it works
The loop
Assemble a prompt with live data. Pipe it to your agent. Repeat.
Assemble
Run commands, gather live data, inject results into your prompt template.
Execute
The assembled prompt pipes to your agent. It reads, works, commits, exits.
Repeat
Back to step 1. Fresh context, latest code. No drift, no decay.
One file. That's it.
A ralph is just
a Markdown file
No config files. No scaffolding. No dashboard.
Create a RALPH.md and you have an autonomous agent loop.
- agent — which CLI agent to run
- commands — live data injected into each iteration
- The prompt body — what the agent sees every loop
---
agent: claude -p --dangerously-skip-permissions
commands:
- name: tests
run: uv run pytest --tb=short
---
You are an autonomous coding agent running in a loop.
Test results from the last run:
{{ commands.tests }}
Fix any failing tests. If all tests pass,
pick the next task from TODO.md and implement it.Why loops beat single runs
Single runs decay.
Loops compound.
A single agent conversation fills up, slows down, and loses the plot. Loops don't.
Fresh context every iteration
No conversation bloat. The agent reads the actual codebase every loop — not a stale 200-message thread.
Self-healing feedback
When something breaks, the failure output feeds into the next iteration. The agent sees what went wrong and fixes it.
Steer while it runs
The prompt is re-read every iteration. Edit it while the loop runs and the agent follows your new direction next cycle.
Progress lives in git
Every iteration commits. If something goes sideways, git log shows exactly what happened.
Agent-agnostic by design
Any agent that reads stdin.
Ralphify is a harness, not an agent. Pipe to whatever you want.
Two minutes. Seriously.
Get Ralph running
Three commands. Zero config. Your agent starts shipping.
uv tool install ralphify
ralph init my-ralph
ralph run my-ralph
Also available via pip install ralphify or pipx install ralphify
FAQ
What AI coding agents work with Ralphify?
Any agent that accepts prompts via stdin or CLI arguments — Claude Code, Aider, Codex CLI, and others. Ralphify is agent-agnostic by design. See supported agents.
What is a ralph loop?
An autonomous AI coding cycle: assemble a prompt, pipe it to an agent, let the agent work, repeat. Each iteration starts fresh — no stale context, no accumulated conversation history.
How is Ralphify different from running my AI coding agent once?
A single agent run fills up its context window, slows down, and loses focus. Ralphify resets every iteration — fresh codebase, fresh data, fresh prompt. You steer, the loop executes.
Does Ralphify require a web UI or JavaScript runtime?
No. It's a pure CLI tool. Runs in your terminal, works over SSH, no browser needed.
What is agent FOMO?
That nagging feeling that everyone else has an AI coding agent shipping code right now — and you don't. Ralph fixes this. Set it up once, and your agent runs autonomously while you're away.
What is an AI agent validation harness?
A validation harness wraps an AI coding agent with automated checks — tests, linters, type checkers — that run after every agent action. Ralphify is a minimal harness that validates each iteration and feeds failures back to the agent, creating a self-correcting loop.
Stop stressing.
Start shipping.
uv tool install ralphify && ralph init my-ralph && ralph run my-ralph