FousAutomating a website with no API

How to automate a website that has no API

There are five practical ways to automate software that has no API: a vendor API, file exports, RPA, an AI browser agent, or a learned web method. The right one depends on whether the job repeats, who owns the site and who fixes it when the site changes.

Updated November 9, 2026

The five approaches compared

Approach	How it works	Logins	What breaks it	Who fixes it
Vendor or partner API	The software’s owner gives you a documented API or an EDI connection.	API keys or OAuth from the vendor.	Version changes, usually announced.	Your team, using the vendor’s notes.
Exports and imports	Someone exports a CSV from one system and imports it into another, sometimes on a timer.	A person signs in, or a service account where one exists.	Renamed columns and missed exports.	Whoever owns the file.
RPA and scripted browsers (UiPath, Power Automate, Playwright)	A bot or script repeats recorded clicks or follows page selectors.	Stored credentials in the bot; two-step codes take extra work.	Layout and selector changes.	The people who maintain the bots.
AI browser agents (Skyvern, Stagehand, Browser Use)	An AI model reads the page and chooses each step in a real browser. Some products can replay cached steps.	Depends on the product.	Misread pages, and cached steps that go stale.	The agent adapts; someone reviews the results.
Learned web methods (Fous)	AI learns the site once. Each run repeats the steps it learned, with no AI model choosing them.	A sealed login with two-step sign-in, and sessions renewed automatically.	Site changes, which trigger a repair.	Fous, in place and at no cost.

Ask for an API first

If the software’s owner offers an API or an EDI connection and will give you access, use it. It is the most stable option, and the only one the owner supports. The cost is time: partner programs and EDI onboarding depend on the other company’s schedule, and many portals don’t offer either.

Use exports when the data can wait

A scheduled CSV is simple and easy to audit, which suits data that changes daily at most. It stops working as an approach once you need a single record on demand, or once a person has to sign in to produce every file.

Use RPA for stable screens you control

RPA tools and browser scripts work well on internal systems whose screens rarely change. They are harder to keep running on sites someone else controls, because any layout change can stop a bot until an engineer updates it.

Use an AI browser agent for varied tasks

An agent can take a task on a site it has never seen, with no setup. That makes it a good fit for one-off or changing tasks. For the same job repeated all day, you are running a browser and possibly a model for each run, and answers can differ between runs. Our comparison of browser agents and methods covers this in more detail, including how cached replays change the picture.

Use learned methods for repeated jobs on known sites

A learned method fits work that repeats with different inputs: order status, invoice status, a registry lookup. Fous builds one in about a minute. Your systems get a typed contract, portal sign-in is handled, and repairs happen without your team.

How to choose

If the owner will give you an API or EDI in the time you have, use it.
If the data can wait a day and a file is available, use an export.
If the task is different every time, use a browser agent.
If the screens are internal and stable and you already run RPA, keep using it.
If the same job repeats on a site someone else controls, use a learned method.

Whichever you choose, check that the site’s terms allow automated access and that the account you use has permission to do the work.

Frequently asked questions

Yes. Tools like Fous learn what a website does for a specific task and expose it as an API call with a defined input and output.

Yes, with an approach that keeps sessions. In Fous you sign in once through the site’s own flow, with emailed codes, sign-in links or phone approvals, and Fous renews the session after that.

An API from the owner, when you can get one. Without one, reliability comes down to what happens when the site changes, so compare how each option detects a change and who repairs it.

RPA suits stable, internal screens with fixed steps. A browser agent suits varied tasks on unfamiliar sites. Neither is built around a typed API contract for your systems to call.

Try it on a site you use

Describe one job in Studio and see the method it builds, with 100 credits a month and no credit card, or book a demo and bring the portals your team works in.