How it works

A loop you can trust with your product.

Four steps on autopilot — and a fifth mode for teams who bring their own developers.

01

Connect your tools

Sign in with GitHub and connect the monitoring you already have — Sentry for crashes, Amplitude for funnels, Microsoft Clarity for rage-clicks. There's nothing to install in your product and no code to write. From that moment, the loop is watching.

SentryAmplitudeClarity→ watching for signals
Connected sources

02

Signals become tasks

When something's wrong — or an opportunity shows up — Ratchet frames it as a task in plain language: what happened, who it affects, why it matters. Tasks are scored and ordered, so the board always answers one question: what should improve next?

SentryHigh priority

Checkout crash when the cart empties mid-payment

214 users hit this in the last 7 days

A framed task

03

It builds and tests

The agent writes the change in a safe copy of your code, then has to prove it: your product's own tests run, and the change deploys to an isolated preview. Fail either check and it goes back to work — you never see half-done proposals.

  • ✓ change built in a safe copy
  • ✓ 128 tests passed
  • ✓ preview verified
  • → ready for your review
Verification before proposal

04

You approve, it ships

You get the before-and-after in plain language, with the test results attached. Approve from your desk or your phone. Decline and it's simply set aside. Either way, nothing reaches your customers without your explicit yes.

Proposal · tested ✓

Fix checkout crash on empty cart

ApproveReview
The approval gate

05 · Management mode

Have developers? Ratchet manages instead of builds.

Flip a product to management mode and the same loop drives your human team — Ratchet does the product-manager work, your developers do the code.

  • Writes the spec and acceptance criteria for every task
  • Files each task as a GitHub Issue for your team, and syncs when it closes
  • Tracks progress against the goals you set
  • Drafts stakeholder updates so reporting writes itself

Under the hood

For the engineer being asked “is this safe?” — the precise version.

Isolated workspaces
the agent works in a safe copy of your repository — never your live code
Preview environments
every change runs in a throwaway deployment before you see it
Your test suite
your repository's own tests must pass on every build
Bring-your-own key
prompts and code go straight to Anthropic on your key — no middleman
Approval gate
no change merges without an explicit human yes

Stop babysitting your backlog.

Free plan · no card required · your own AI key

Start free