Claude agents, on your GitHub repos

Your backlog, shipped as pull requests.

Write a ticket. A Claude agent builds it in a cloud sandbox or on your machine and opens a GitHub pull request. You review and merge.

Log inRead the docs

Runs on your own Anthropic API key.

#483 · Backlog
Where the agent runs

Edit the ticket and send it. This demo runs in your browser; nothing is sent anywhere.

How it works

One ticket, start to finish

A typical bug fix, from the moment you write it down to the moment it's on main. You spend about two minutes of it.

  1. 09:02

    You write the card

    Describe the bug in plain words and move it to Ready for Agent.

  2. 09:03

    An agent picks it up

    Claude starts in a fresh sandbox, or on your machine, with a clean clone of the repo.

  3. 09:14

    A pull request opens

    Changes, a summary and test results land on GitHub as a normal PR.

  4. 09:20

    You merge it

    Read the diff, click Accept & Merge. Nothing reaches main before that.

Review

Review the diff, not the chat log

Every run ends as a real GitHub pull request with the agent's summary and test results. Review it in RelayDev or on GitHub, then merge from either.

Pull request #482

Retry failed Stripe webhooks

relay/task-482 → main · +111 −9

Needs Review

Agent summary

Wrapped the webhook handler in a retry with exponential backoff and added a dead-letter table for events that fail five times.

  • src/app/api/stripe/route.ts+41 −9
  • src/lib/stripe/retry.ts+58 −0
  • supabase/migrations/0042_dead_letters.sql+12 −0
48 tests passedTypecheck
Request changesAccept & Merge

Cloud or local

Run it in the cloud or on your laptop

Pick per card. Ready for Agent runs in a cloud sandbox with nothing to install. Ready for Local hands the task to the desktop app and your own Claude Code, git and toolchain.

Ready for Agent
#481

Rate-limit the public webhooks endpoint

Vercel Sandbox microVM

Ready for Local
#479

Upgrade the iOS build to Xcode 18

Desktop app on your machine

Usage

Your key, your token bill

Agents run on your own Anthropic API key. RelayDev shows usage per project and lets you set budgets, but never meters or marks up tokens.

Tokens this month

  • checkout-service1.2M / 1.5M
  • marketing-site420k / 1.5M
  • mobile-api160k / 1.5M

  • Results from your CI

    Upload JUnit reports from any CI and each result lands next to its test case.

  • Main stays yours

    Agents can't push to your default branch. Every change arrives as a pull request.

  • One sandbox per task

    Cloud tasks each run in their own Vercel Sandbox microVM, so runs never collide.

Test Suite

Every behaviour of your app, in one list

An agent maps your repository into views and modules. Each test case is a sentence about what the app should do, with its latest result next to it, so you see what's covered and what isn't.

Checkout

src/app/checkout/page.tsx

Implement 2 missing tests
8Passed1Failed1Being written2Untested

  • Applies a discount code

    Entering a valid code updates the total before payment.

    Passed

  • Rejects an expired card

    Shows the card error inline and keeps the form filled in.

    Failed

  • Charges tax by country

    An Austrian address adds 20% VAT to the order.

    Being written

  • Guest checkout

    A visitor can pay without an account and gets a receipt by email.

    Untested

Write tests in words

Describe it, an agent writes the test

Add a test case the way you'd explain it to a colleague. Pick the missing ones and an agent implements them as unit or UI tests, in one pull request or one each. Checks a machine can't do stay on a manual checklist.

New test case · Checkout

Guest checkout

A visitor can pay without an account. After payment they see the order number and get a receipt by email.

Automated UIHigh

Test Suite pull request #212

e2e/checkout/guest.spec.ts · +64

Waiting for merge

Suggestions

Test ideas with every pull request

When an agent finishes a ticket, it proposes test cases for what it just changed. Accept the ones that matter and they join the Test Suite, ready to implement. Dismiss the rest.

Pull request #482

Retry failed Stripe webhooks

3 suggested test cases

  • Retries a failed webhook three timesHigh
  • Moves an event to dead letters after five failuresHigh
  • Ignores a duplicate event IDMedium
Accepted cases join the Test Suite as Untested, ready to implement.

Pricing

Pay for the platform, not the tokens

Plans cover projects and cloud hours. Tokens go on your own Anthropic bill, and local runs are always unlimited.

Local

Run agents on your own Mac, follow them on the web.

€2.99/month

or €23.99 a year, 4 months free

  • 3 projects
  • Unlimited local runs in the desktop app
  • Watch runs, review and merge from the web
  • Built-in Test Suite
  • Token budgets and usage dashboards
  • No cloud runs: needs the desktop app
Start with Local

Solo

Most popular

For one developer shipping side projects.

€11.99/month

or €119.99 a year, 2 months free

  • 14-day free trial with 3 cloud hours, no card needed
  • 3 projects
  • Unlimited local runs
  • 20 cloud hours a month
  • Runs up to 45 minutes each
  • Built-in Test Suite
  • Token budgets and usage dashboards
  • Extra cloud hours at €0.50 each
Start 14-day free trial

Pro

For developers running agents all day.

€29.99/month

or €299.99 a year, 2 months free

  • Everything in Solo
  • 10 projects
  • 60 cloud hours a month
  • 3 cloud runs at the same time
  • Runs up to 45 minutes each
  • Extra cloud hours at €0.50 each
Get started

FAQ

Questions

Do I need my own Anthropic account?

Yes. Agents run on your own Anthropic API key, added once in Account settings. RelayDev encrypts it at rest and never charges for tokens.

What's the difference between cloud and local runs?

Cloud runs happen in an isolated Vercel Sandbox microVM. Local runs happen on your own machine through the desktop app, using your existing git and Claude Code install.

What counts as a cloud hour?

The time a cloud sandbox runs for one of your tasks, from start to stop, counted by the minute. Hours reset every month on your billing date. Local runs never count, on any plan.

What happens when the trial ends?

You pick a plan. Local keeps every run on your own machine for €2.99 a month; Solo and Pro add cloud hours. Until you choose, your projects, tasks and pull requests stay as they are, but no new runs start. Nothing is deleted.

Can agents merge code without me?

No. Every run opens a pull request. Nothing reaches your default branch until you review the diff and click Accept & Merge.

Which repositories can agents see?

Only the ones you grant the RelayDev GitHub App. You can change that list on GitHub at any time.

Send your first ticket today.

Connect a repository, add your Anthropic key and write one card. Solo is free for 14 days, and every plan includes the desktop app.

Log inSetup guide