# Deck Browser Agents: Automate any website your product needs

> Deck's browser agents log in, click through pages, and fill out forms on sites never built for automation, then send structured data back to your product.

Source: https://deck.co/agents

Browser Agents

Deck’s browser agents log in, click through pages, and fill out forms on sites that were never built for automation. You define the task, and Deck sends structured data back to your product.

[Get Started](https://console.deck.co/)

[Talk to Sales](https://deck.co/contact)

## Agents that use the web the way a person does

Most of the software your users rely on never shipped an API. Utility portals, insurance carriers, payroll systems, and government sites only work through a browser. A browser agent opens that website, reads the page, and has access to the same data a person would.

### Fetch

Pull bills from PG&E, claims history from Aetna, or invoices from QuickBooks.

[Explore bill fetch](https://deck.co/use-cases/bill-fetch)

### Submit

Fill out applications, update records, or file claims on portals that only accept a form.

[Explore product integrations](https://deck.co/use-cases/product-integrations)

### Pay

Process rent through ClickPay or pay a utility bill for your user.

[Explore agentic payments](https://deck.co/use-cases/agentic-payments)

## One task, every website

Write a task like “download this month’s bill” once, and it runs on every utility website you’ve connected, whether that’s PG&E, Duke Energy, or a small municipal portal. You don’t build or maintain a separate integration for each site.

[See how one task works across sites](https://docs.deck.co/how-deck-works#deterministic-output)

## Keeps working when websites change

Most automation breaks when a website changes. Deck’s agent harness remembers what worked on a site before, and recovers on its own when something unexpected shows up. If a page loads slowly or a layout shifts, the agent finds another way through. You set the objective, and the agent deals with whatever the site throws at it along the way.

[Learn about the agent harness](https://docs.deck.co/concepts/tasks#writing-the-prompt)

## See every step of every task

You define the output schema, and every run returns JSON that matches it, ready to drop into your product. Deck takes a screenshot every 5 seconds and logs the agent’s reasoning, so when something looks off, you can see exactly what happened. You can also watch agents work live in the Deck Console.

[Learn about agent observability](https://docs.deck.co/platform/agent-observability)

## Hosted infrastructure

Each task runs on its own isolated computer, with a full desktop environment that Deck hosts and scales for you. You don’t manage browsers, proxies, or servers. A machine starts instantly and shuts down when the task is done, so running one task works the same as running thousands.

*Diagram: your product sends Deck one task or thousands. Deck admits each task into its agent pool, starting an isolated machine for each one and shutting it down when the task is done. A callout follows one task execution: the task for its source, a browser with a stealth profile, and an auth status that changes to logged in, moving from queued to running to completing to completed. Completed tasks report back to your product.*

## If a person can do it in a browser, Deck can too

Start building today or talk to us about running at scale.

[Get Started](https://console.deck.co/) · [Talk to Sales](https://deck.co/contact)

## More from Deck

- [Deck home](https://deck.co/)
- [Auth](https://deck.co/auth)
- [Tasks & Automation](https://deck.co/tasks-and-automation)
- [Observability](https://deck.co/observability)
- [Pricing](https://deck.co/pricing)
- [Security](https://deck.co/security)
- [Customers](https://deck.co/customers)

Docs: https://docs.deck.co · Site index for agents: https://deck.co/llms.txt