PRD-driven loop
Generate a PRD and task list from raw requirements so the loop has a durable source of truth instead of a single prompt.
Ralph Loop
_____ _ _____ _ _
| __ \ /\ | | | __ \| | | |
| |__) | / \ | | | |__) | |__| |
| _ / / /\ \ | | | ___/| __ |
| | \ \ / ____ \| |____| | | | | |
|_| \_\/_/ \_\______|_| |_| |_|
A long-running AI agent loop that can code for days.
Ralph iterates through a task list, runs your favorite agentic CLI in Docker Sandboxes, and keeps working until the job is done.
$ npx @pageai/ralph-loop█
agent-loop.exe
// live-ish terminal playback
Point it at a task list. Let it loop.
ralph loop sessionpid: long-running
$ sbx run --name ralph-loop claude .
ralph: docker sandbox online
[001] task: create PRD + task list ✓
[002] task: implement feature slice ✓
[003] task: run tests and typecheck ✓
ralph: loop complete, ready for review █
// start here
Watch the walkthrough
▶ ralph wiggum loop demo
// features
Built for unattended agent work
Ralph is a hackable, long-running loop. Point it at a task list, walk away, and come back to a stack of commits.
Task lookup table
Per-task specs with detailed steps. Scales to hundreds of tasks without losing context between iterations.
Docker Sandboxes per agent
Deterministic ralph-<agent>-<dir>-<hash> sandboxes. Isolated, reusable, and stopped cleanly on exit.
Multi-agent ready
Claude, Codex, Cursor, Copilot, Gemini, opencode out of the box. Pass agent flags through the same script.
Live observability
Step detection, stream preview, screenshot capture, full history logs, and per-iteration timing while it runs.
Steerable mid-flight
Edit .agent/STEERING.md while the loop is running. Ralph reads it each iteration and prioritizes critical work.
From the blog
All posts ->
How to Inspect and Debug Inside an AI Agent Sandbox
A sandbox is not a black box. Here is how to find the running microVM, shell into it with sbx exec, reproduce the failure by hand, and read the network log.
Network Policies for AI Agent Sandboxes
A sandboxed agent should start with no outbound network and earn each domain it reaches. Here is how to allowlist exactly what a task needs with sbx policy, so package installs work and exfiltration does not.
Running Agents in YOLO Mode (--dangerously-skip-permissions) Safely
Bypass-permissions mode hands an agent full shell access with no prompts. That is reckless on your host and a non-event inside a Docker Sandbox. Here is how to run YOLO mode when the blast radius is contained.