For developers who run Claude Code, Codex — or both

Your AI coding team runs the night shift.

Schedule Claude Code and Codex on the subscriptions you already pay for — each driven to its full power, not a dumbed-down generic agent. They run unattended within a budget cap; you triage everything from one inbox.

Free while in beta. You already pay for your coding agent — ratchetAI is the night shift on top. Flat pricing at launch, never per-run.

NO NEW SEAT TO BUY
CREDENTIALS STAY LOCAL
SPENDS ONLY TO CAP
HOW IT WORKS

Configure once. It runs itself — and stops when it shouldn't.

workflow.diff— the shift in one read
-chat with an agent, watch it think, repeat
+configure automations → they run themselves → you observe & steer
01

SCHEDULE

An objective, a cadence, a budget cap.

02

RUN

Your local, already-authenticated CLI runs non-interactively.

03

SUMMARIZE

A plain-language summary lands in your inbox.

04

DECIDE

It runs the next step — or escalates to you.

Credentials never leave your machine. Deterministic guardrails — budget caps, repeated failures, risky steps — pause the loop before it does damage and hand you a one-line steer. Free while in beta.
WHY RATCHETAI

It runs your Claude Code and Codex — each at its max, and together.

Other tools wrap one agent. ratchetAI reads both of your setups, drives each engine to its individual best, and puts them on the same job — so the run isn't limited to one model's blind spots. Pay for only one CLI? It still works — the second engine is the upgrade, not a requirement.

KNOWS BOTH SETUPS
EACH ENGINE AT ITS MAX
CHAINS THEM ON ONE GOAL
UNATTENDED + BUDGETED
01

KNOWS BOTH SETUPS

It detects each engine's tools, MCP servers, and permission posture — Claude and Codex — and tailors every run to what's actually on your machine.

02

EACH AT ITS MAX

Expert per-engine prompting — parallel subagents, your skills and MCPs, plan→verify, ship-to-live — so each CLI runs at the level a power user gets interactively, not a watered-down default.

03

THEY CHECK EACH OTHER

One engine builds; the other independently verifies before it's accepted — two engines cross-checking beats one grading itself. Rolling out in beta.

No single-engine tool does this. Neither does a raw claude -p cron. The defensible part isn't running an agent unattended — it's understanding your Claude Code and Codex deeply enough to run each at its best and put them on the same goal together.
PROOF

We pointed it at itself — and it shipped.

To test ratchetAI we gave it a deliberately vague goal — "build something more trustworthy and simpler to set up than OpenClaw or Hermes Agent; you decide what it is" — and let it run unattended. Across two chained sessions it chose a product, built it, and verified it, then stopped on its own when the acceptance criteria were met — not on a timer. This is our own dogfood, run in public.

2 CHAINED SESSIONS
48 / 48 TESTS PASSING
STOPPED ON VERIFIED DONE
BUILT IN PUBLIC
What it built: a small, tested TypeScript v1 — a tamper-evident execution ledger for AI-agent runs — scaffolded in the first session, then verified (build + tests + README) in the second, with a plain-language summary landing in the inbox at each step. No human in the loop between the goal and the verified result.
QUICKSTART

From zero to your first night shift.

Sign up free, connect your machine with one command, and describe your first automation in plain language. It runs tonight.

~/your-repo — zsh Requires: Node 22+, and the claude (or codex) CLI installed & authenticated
connect this machine — it drives your local claude / codex CLI
$ npx ratchetai-runner connect --url https://ratchet-control-plane.wandering-voice-f671.workers.dev --token rt_… --cwd ~/code
Mint your token in the app (Runners → New token), then describe automations in plain language — the AI configures each one. Free while in beta; flat pricing at launch, never per-run.

Want to read it first? Browse the source on GitHub — it's built in public.