If your app runs on your laptop, it runs in YourBox. Each agent gets a cloud box with your whole local setup running, hot reload, its own database, and logins that work.
Works with Claude Code, Codex CLI, Cursor CLI, and OpenCode.
Agents do their best work when they can run the app and check what they changed. One copy fits on a laptop. Several copies fight over ports, databases, and memory.
Stripe and Ramp each built this for themselves, and both have written about how it works.
Stripe's agents run on the same pre-warmed cloud devboxes its engineers use, one box per task. An engineer can start several from Slack and come back to finished pull requests.
Read Stripe's post (opens in a new tab)Every Ramp session gets a full environment with Postgres, Temporal, and the rest of the stack. Agents check their work with screenshots and live previews, and people outside engineering get the same setup an engineer would.
Read Ramp's post (opens in a new tab)Both were built by dedicated platform teams. YourBox gives your team the same setup without one.
See how it worksStripe and Ramp are not affiliated with YourBox. Figures come from their own engineering blogs.
Set it up once. After that, starting a box is one command, and your agents can do it themselves.
app: acme-web services: web: npm run dev # :3000, hot reload api: go run ./cmd/api # :8080 worker: npm run worker postgres: postgres:16 redis: redis:7 seed: scripts/seed.ts # synthetic accounts url: $BOX_URL # replaces localhost:3000 check: npm run smoke # sign in, pay
The recipe lives in your repo. When the README is out of date, the recipe still reflects what boots.
The same link your agent tests is the one you and your reviewers open.
Reviewers click a link instead of pulling the branch and fighting their own setup.
Getting an app to start is the easy half. A copy is only useful when you can log in, see realistic data, and finish the flow you're changing.
Every box gets a fresh database seeded from your recipe: test accounts, plans, and feature flags that make the app usable. Nothing comes from production.
| users | plan | flags | box |
|---|---|---|---|
| ada@acme.test | Team | new-checkout | checkout-flow |
| lin@acme.test | Trial, day 13 | apple-pay | checkout-flow |
| sam@acme.test | Enterprise | sso | checkout-flow |
| kai@acme.test | Past due | none | checkout-flow |
Each box gets a real URL, so the settings that expected localhost come along: app URLs, cookies, login redirects like Google sign-in, and webhooks.
Agents open the running app, click through the flow they changed, and attach screenshots to the pull request.
Web, API, workers, queues, and databases, across one repo or several.
The recipe is rechecked whenever main changes. When something breaks it, you get a proposed fix.
Any agent that runs in a terminal can start, use, and stop boxes. The same commands work from the CLI, through MCP, or over the API, so orchestrators can give every task its own box.
$ yourbox up checkout-flow Starting from recipe acme-web web api worker postgres redis Ready at https://checkout-flow.acme.yourbox.dev $ yourbox exec checkout-flow -- npm test $ yourbox logs checkout-flow api $ yourbox down checkout-flow
// .mcp.json in your repo { "mcpServers": { "yourbox": { "command": "yourbox", "args": ["mcp"] } } }
$ curl https://api.yourbox.ai/v1/boxes \ -H "Authorization: Bearer $YOURBOX_TOKEN" \ -d '{"recipe": "acme-web", "branch": "checkout-flow"}' { "id": "box_7f2c", "status": "ready", "url": "https://checkout-flow.acme.yourbox.dev" }
Boxes are for building and testing, so they never need anything from production.
Questions from your security team? Email security@yourbox.ai and we'll answer them directly.
Boxes run on synthetic seed data from your recipe. Nothing is copied from your production databases.
Payment keys run in test mode, and sign-in uses a development-only OAuth client.
App secrets come from your secrets store and stay separate from your agents' credentials.
Every box link requires your team's sign-in. Nothing is public unless you choose to share it.
Boxes run in our cloud by default. For teams whose code must stay in their own AWS account, we're setting that up with design partners now.
Anything else, email hello@yourbox.ai.
Early access is free for design partners. We'll share pricing well before anything changes, and nothing is charged without your agreement.
If it runs on a laptop with a README or setup scripts, we can usually get it running. Apps that already use Docker Compose are the quickest to set up.
Those give you a machine. YourBox gives you your app running on it at its own link, with seeded data and working logins, and keeps it working as your code changes.
Yes. Anyone on your team can open a box's link in a browser to see a change running, with nothing to install.
We're working with a small group of design partners. We'll get your app running for your agents, free, in exchange for 30 minutes of feedback.
This branch is running in its own box. Sign in with your team account to try it.
https://checkout-flow.acme.yourbox.dev