The jankiest of forms. The internal tool built by Jeff that your company relies on for basically everything. A government website that looks like it was designed in 2002 but has crucial information for your business. What do all of these have in common? They have no connector, no way for your agents to get to them. That is, until now.
Now when you ask your agent to do something where Gumloop doesn’t have a native connector, it opens up a browser and uses the site like you would. Here are three examples of how to use this new superpower.
I’ve got an agent here that’s connected to Slack with browser use on, and I want this agent to monitor building permits in Sacramento, which are posted every morning, and send me a Slack DM with that day’s list. I can simply ask, “Hey friend, can you get today’s Sacramento building permits and DM them to me on Slack?” Remember to always be nice to your agents.
Since this agent has no connector to retrieve that information, it’s going to open up a browser, navigate to the page where the information lives, and try to accomplish the task. The agent looks through the elements on the page and figures out which ones it can use, just like we would. It can grab information from the page and input data into fields. It’s just browsing the web.
Now it’s found today’s permits, extracted them from the page, and used the existing Slack connector to send them to me. Next I can say, “Save what you’ve learned in navigating this page as a skill, and send me a Slack DM with that day’s permits every morning at 9 a.m.”
Browser use is no different from other agent processes. We can save the steps as a skill and have it run on a schedule or on a trigger, like we just did.
A quick note on credentials. If the agent encounters a login page, it’ll loop you in to input login details. Logins can be securely saved as secrets on the agent, or you can give your agent access to a 1Password vault so it can grab logins for you automatically.
Another example. Say you need to copy and paste details from an email into various portals, like getting an insurance quote or completing manual steps in a workflow. An agent can now do that manual work for you and input details into a form.
Here’s a request from a customer with their details that I would normally copy and paste into multiple forms. Instead, I can forward it to my agent. The agent reads the email, reads its skill so it understands which portals it needs to access and how, and starts inputting the information. When the agent is done getting the quotes or whatever else it needs, we get a response directly in our inbox.
One final example. Want to test a process end to end? An agent can take care of that for you. Something like, “Every day at 9 a.m., I want you to test the new customer discount flow on allbirds.com and email me if there are any issues.”
The agent will navigate to the site, find the discount code or figure out how to get it, enter it, and validate that it works by putting items in the cart and applying the code. If that doesn’t work, it emails you. Just like that, we’ve set up a quality assurance workflow.
Now it’s over to you. What’s a site you log into every day to grab some piece of information? A form you’re tired of copying and pasting into? Give your agent a browser, show it once, and let it take over from there.
Browser Use
Browser Use lets your agents open a browser and work with any website, so a missing connector is no longer a dead end.
Connectors cover a lot of ground, but plenty of important work lives somewhere no connector will ever reach. The internal tool one engineer built years ago. A county website that only updates by hand. A partner portal with a form that has to be filled out every single time.
Browser Use closes that gap. When an agent needs something from a site Gumloop has no native connector for, it opens a browser and works through the page the same way a person would.
How it works
With Browser Use turned on, the agent reads the elements on a page and decides which ones matter for the task. From there it can:
- Navigate between pages to find what it needs
- Read and extract information from the page
- Type into fields and fill out forms
- Click through multi-step flows like a checkout
The browser is just one more tool. The agent can grab data from a site, then hand it off to a connector you already use, like posting the results to Slack or replying in email.
Turn one run into a repeatable process
The first time an agent works through a site, it has to figure things out. Once it has, ask it to save what it learned as a skill. Next time it already knows where to go and what to click.
From there, Browser Use behaves like any other agent work. You can put it on a schedule, kick it off from a trigger, or send it work by email.
Handling logins
When the agent hits a login page, it pauses and asks you for credentials instead of guessing. You have two ways to make that smoother over time:
- Secrets: save logins securely on the agent so it can sign in on its own.
- 1Password: give the agent access to a vault so it can pull the right credentials automatically.
Where it shines
- Monitoring sites with no API: check a public records page every morning and send a summary of anything new.
- Data entry across portals: forward a customer email and let the agent copy the details into each form, like collecting insurance quotes.
- End-to-end testing: have the agent walk through a signup or discount flow on a schedule and email you when something breaks.
What to remember
If a person can do it in a browser, your agent probably can too. Start with a site you visit every day or a form you are tired of filling out. Walk the agent through it once, save it as a skill, and let it run from there.
