Build Agents That Run Code.

Runs your handler in a sandbox with a public URL, fires Stripe test events at it, reads the errors, and patches the code until every event returns 200.

For: SaaS developers whose webhook handler silently fails in production.

Takes a Sentry error, writes a minimal script that reproduces it, confirms it crashes the same way, then proposes a fix.

For: On-call engineers staring at a stack trace they can't reproduce.Sentry

Pulls every code block from a docs page, runs each one, and reports which are broken with a suggested fix.

For: DevRel teams whose docs code examples are broken and nobody knows.

Turns a brief and a logo into a page, shares a live link that sleeps until the client opens it, and hands over the code once the client pays a deposit through Stripe.

For: Freelancers who lose clients waiting days for a first draft.

Writes and runs Python against your Stripe test data, in a sandbox whose network can only reach api.stripe.com.

For: Founders who want MRR, churn and cohort charts but have no data team.

Takes both CSVs, writes the matching logic, flags mismatches, and exports a clean report.

For: Small-business accountants matching Stripe payouts to bank statements by hand.

Runs each submission against hidden tests with no internet, explains the failures, and the course pays per batch.

For: TAs grading 200 Python submissions every week.

Parses the export, calculates hours times rate, and creates the Stripe invoice.

For: Freelancers turning Toggl or Clockify exports into invoices.

Installs the package with the network locked down, logs what it tries to reach, and gives a safe or risky verdict.

For: Developers adding unfamiliar npm or pip packages to a project.

Opens the link in a throwaway Chrome, screenshots each redirect, explains the scam, then deletes its own sandbox.

For: Small companies with no security team, getting suspicious emails.

Writes and runs refund code that can only reach Stripe, with a capped amount and a human approval step.

For: Support teams who want AI to handle refunds without the risk.

Navigates your site keyboard-only in real Chrome, screenshots the problems, and writes a WCAG report with fixes.

For: Companies that must meet accessibility rules, such as the EU Accessibility Act.

Signs up for your product as a stranger and notes every confusing step with screenshots, while you watch it live.

For: PMs who never see their own signup flow the way a new user does.

Clicks through the app and regenerates every screenshot for the docs.

For: Teams whose docs screenshots go stale after every UI change.

Wakes daily, pulls Stripe and PostHog data, writes the analysis, emails three charts, and goes back to sleep.

For: Founders checking five dashboards every morning.PostHog

Forward receipts to an email address, and the agent extracts the amounts and keeps a running expense sheet.

For: Freelancers dreading expense reports at tax time.AgentMail

Pause your Python session and share it. A teammate forks it with every variable still in memory.

For: Data teams handing off half-finished notebooks.

An alert wakes the sandbox, the agent runs health checks and log queries, then emails a diagnosis.

For: Small teams with no on-call rotation.

Give the agent one tool, execute, instead of twenty. It writes a single script against the Stripe API in test mode and runs it in the sandbox.

Demo: The same task done both ways, with the token count beside each.Stripe test mode

From a one-line spec, it adds a database and auth with Stripe Projects, writes the app, tests it against the real services, and deploys it.

Demo: An empty directory becomes a live URL, and no dashboard gets opened.NeonClerkCreateOS

It provisions Neon, Turso and PlanetScale through Stripe Projects, benchmarks your workload on all three, keeps the winner and removes the rest.

Demo: A live latency chart, then the losing databases deleted.NeonTursoPlanetScale

Put a sandbox behind an HTTP 402. Other agents pay per execution with Stripe's Machine Payments Protocol, all in test mode.

Demo: An agent hits the 402, pays, gets its output, and the earnings tick up.Stripe MPP

The agent edits the code, runs a 60-second experiment, scores it, and keeps or reverts the change. Then it does it again, forty times.

Demo: A metric climbing on screen, and the git log of every hypothesis it tried.OpenRouter

When it hits a gap, it writes a tool, tests it in the sandbox, and saves it as a skill. The next session starts with it.

Demo: The same task run twice, the second time much faster.Supermemory

It turns a domain into a scored benchmark, with tasks and a checker that runs in the sandbox, then puts three models through it.

Demo: A leaderboard for a benchmark that didn't exist that morning.OpenRouter

Set up a repo with a failing test, pause, and fork it into four identical copies. Each tries a different fix, and the tests pick the winner.

Demo: Four tiles racing until one turns green.GitLab

A vulnerable app in one sandbox, an attacker agent in another. It has to land a working exploit, write the patch, then run the exploit again.

Demo: The exploit works, the patch lands, the exploit fails.

A planner hands tasks to coders, each in its own sandbox. They claim work, push fixes, and the test suite keeps score.

Demo: A pass-rate bar filling up.GitLabOpenRouter

In a desktop sandbox with Chrome, the agent clicks through an app that has no API, then writes a Playwright script so the next run needs no clicking.

Demo: The audience watches it work, live, over VNC in the browser.

From a Sentry error, it boots the real stack, writes a failing reproduction, fixes it, and opens a pull request with the evidence.

Demo: A red test, a green test, a pull request.SentryGitLab