Getting started

Sign up at sprawl.to/sign-up — every account starts with $3 of usage, no card required and nothing to cancel. There is no subscription: each run is priced for the computing work it actually used, typically 6¢ to 16¢, and drawn from your balance. That first $3 covers roughly 30 automation runs (details on the pricing page).

Connect GitHub and Vercel and Sprawl surveys what you've shipped — your repos, deployments, and docs — then proposes the operation to run it. You pick which projects to operationalize. Or skip the survey and describe any operation from scratch, the way you'd brief a colleague (“triage my client inbox and draft replies every morning”) — both paths work.

Either way, each operation is its own container, cleanly walled off: its own connected accounts (a different Gmail, Slack, or X per operation) and its own fleet of agents. Running a business and a side project? Five separate clients? Nothing bleeds across.

The builder shows the whole operation as one blueprint: the agents, their schedules, the goals they work toward, and the accounts each agent needs — each agent on the model that fits its job, from a catalog of 200+ models across OpenAI, Anthropic, Google, and more. Approving it provisions everything at once; nothing is created until you say so. Want changes? Say so and the blueprint updates before anything is built.

After provisioning, work through the connect checklist. Each account you link flips to ✓ and brings its agents online — one sign-in can activate several agents. Skip anything you're not ready for and it stays parked until you are. Connect an app as many times as you need (a different Gmail, Slack, or X per operation); each connection uses that service's own OAuth consent screen, and you can disconnect any tool at any time.

Actions are classified by risk. Routine reads and drafts run without friction; riskier actions — like sending an email on your behalf — are held in your approval inbox until you approve or reject them, and your decision is recorded. The highest-risk category of actions cannot run autonomously at all. If you reject something, Sprawl takes the feedback and adjusts instead of repeating the attempt.

Any task can become a standing automation: schedule it (“every weekday at 8am”) or trigger it on events (“when a new payment lands”). Scheduled and event-triggered runs use the same approval rules and the same pricing as any other run — autonomy never widens what the agent is allowed to do.

You pay for work, in dollars, as it happens: each run costs what the computing work it actually used costs, and comes off your balance. Building and setup in the builder are free on the standard models (up to 50 turns a day) — your balance goes to the work your agents do, not to describing it. Your balance and per-run history live in the billing page, and you can top it up any time from $5; it never expires. Full details on pricing and in the FAQ.