idle
iter0 session0 runnerclaude branch— treeclean uptime00:00:00 socketconnecting
idle

No loop running. Start loop = continuous mode (until stop conditions hit). Run once = single iteration. Pick a runner and (optionally) a command in Loop configuration below — empty command auto-uses a safe fallback so you can test instantly.

claude —
Space pause
Iterations0
Plan tasks— / —
Specs0
Backpressure—
How to use
A 4-step workflow. Each card below maps to one of these steps.
  1. 1

    Point at a repo

    In Loop configuration below, paste the absolute path to the repo you want Ralpharium to watch (or leave the current one). It reads PROMPT.md, AGENTS.md, IMPLEMENTATION_PLAN.md, and specs/*.md from there.

  2. 2

    Pick a runner

    Choose Claude / Codex / Aider / custom and (optionally) override the shell command. Empty command = a safe fallback iteration runs (so you can test without setting anything up).

  3. 3

    Run it

    Run once = a single iteration to test the wiring. Start loop = continuous until max_iterations, a failure (with stop-on-failure), or you click Stop / Panic.

  4. 4

    Watch what changed

    The Iteration timeline shows every pass with status, files changed, commit SHA. Plan health, Backpressure, and Spec coverage update each iteration. For live runner output and per-agent activity, open /ram.

Safety: turn on stop_on_failure, stop_if_no_commit, and stop_if_dirty_before_run for your first real run. Press Space to pause between iterations, Ctrl/⌘+P for panic stop.
Loop configuration
—
Repository path
Folder Ralpharium watches for PROMPT.md, AGENTS.md, IMPLEMENTATION_PLAN.md, specs/, package.json. Stop the loop before switching.
Runner command
Shell command run each iteration. Empty = manual mode. Save changes to apply.
Before running on a real repo
  • Commit or stash your own changes first. A dirty tree is a footgun for any iteration runner.
  • Start with max_iterations = 1. Validate one full pass before letting it loop.
  • Use a throwaway repo for the first real run. Don't hand a brand-new prompt to a production codebase.
  • Enable stop_on_failure. The first failure usually surfaces a misunderstanding worth fixing in PROMPT.md.
  • Enable stop_if_no_commit. An iteration that doesn't produce a commit usually means the runner didn't really do anything.
  • Enable stop_if_dirty_before_run. Refuse to start a new iteration on uncommitted changes.
  • Confirm the runner CLI is authenticated. Run claude --version / codex --version / aider --version first; do one interactive run to set up auth.
  • Watch the RAM page during the first run. The event stream shows process output, exit codes, and blackboard updates in real time.

This panel won't auto-run anything. It's just a checklist before you click Start loop.

Iteration timeline
0 iterations
No iterations yet.

Start the loop, or have your CLI POST /api/iterations when each pass begins and PATCH when it ends.

Plan health
—
No IMPLEMENTATION_PLAN.md.

Add one to the repo root — Ralph reads it as persistent state between iterations.

Backpressure
—
No validation detected.

Add scripts to package.json (test, lint, typecheck, build) or pyproject — Studio auto-detects them.

Spec coverage
—
No specs/ directory.

Create specs/*.md as the source of truth for what Ralph should build.

Runtime — live RAM
—

Blackboard

Volatile slots.

Loop mode, current task, last error — populated when a runner is active.

Memory pressure

prompt context— B
repo scan— B
events buffer— B
runner RSS—

Event stream 0 events

No events yet.

Loop start, iteration boundaries, scratchpad notes, and checkpoints stream here in real time.

Scratchpad

  • No notes yet — pin a thought to remember it across the session.

Checkpoints

  • Snapshot the repo + plan to compare later.
Prompt & guardrails
—
PROMPT.mdmissing
—
AGENTS.mdmissing
—

Suggestions

Looking good.

No failure patterns yet. Suggestions appear after a few iterations.