Turn any repetitive task into an API. We build an agent that reverse-engineers the network requests to create APIs/tools for your tasks. This is a new paradigm for browser agents, making them reliable, fast, and cheap.

Who wants to try the beta version?

Good to see Magnus

@grok can you explain in simple terms what it means to “turn any repetitive task into an API” by reverse-engineering browser network requests? How would an agent like this actually work in practice, and what kinds of everyday web tasks is it best suited for?

Browser agents that can reverse-engineer network requests are a game-changer! This new paradigm eliminates the need for manual API documentation. Making automation fast, reliable and cheap - exactly what developers need!

@grok how does this compete with $path, maestro, and their RPA. Please be clear and in depth

Even this guy can automate his flows now. Wondering where he gets annoyed in the browser

Thats neat!

Classic trade off between determinism and agency. When you try to hard code at this level the slightest api or website change will trash the entire script 🗑️

But then you just run the skill generation agent once

I've just tested it. First attempt and worked perfectly. This is great @mamagnus00 ;)

Lets create some tools for salesforce

Got a couple of great ideas, I’ll send a short meeting invite for a catch up!

Super cool!

@DubbingAgent dub in Japanese

Excited to try it out

Would love your feedback

When you say Internet, you mean only the DOM in the browser? WhatsApp (WA) Web works in the browser (but not via the DOM). On top of that Meta injects js, to listen for automation? WA allows you to easily build msg list of 256 recipients but won't let you monitor open rates!!!

help me w linkedin api plsssss

Can I test it ?

Yes would love your feedback

Will it works with legacy banks websites? And when I say legacy I really mean it.

Yes. Which page do you have in mind

I want to sign in on my LinkedIn to check my messages, tried to pass my LinkedIn cookies but i'm not sure if i did it correctly, perhaps we could re-use an existing browser with a logged session :)

Will a demo today for that

Nice! Does the skill api support a specific profile as a parameter ? If you have any docs or prefer a demo I appreciate

The agent can make a parameter for the agent cookie

Ok 👌🏻

I want to try

If this instance can be made accessible into an API call. It'll essentially become a tool to create APIs for any services

Will it work for sites requiring login? If so count me in

Yes one parameter will be the session cookie

I have uses cases for this. Would love to beta test

What do you want to try?

Sent dm

Turning UI interactions into API calls is a game changer. Curious about the reliability during complex tasks. What's the accuracy like?

Would screen recording my workflow be good for such tools? Should i put in the effort to store this?

Actually you can just prompt and the agent will do everything. Today we will release a feature to record the workflow

I’m not going to do this, but in this modern era what stops anyone going into Claude and pasting prompt: build an agent that reverse-engineers the network requests to create APIs/tools for your tasks.

@FactsForFarmers Did you try? For me it struggles with network requests + filtering them down + hosting them - just the end2end flow

@jakebrowatzke how does this type of product affect PATH?

@tuffhajj @jakebrowatzke Until the website changes their internal APIs, then we send the agent again and have the new prompt

Love to try it out. Does it support mcp ?

It's like MCP its just an API with parameters.

Interesting. What's the cost?

Right now free, long term much less than 0.01$ per workflow

How do you handle request cookies?

Browser use can go, get the cookie and provide it to the tool

Got it. But then how is your tool faster? Cookies are usually tied to a session so if I rely on the reverse engineered apis I’d still need to get fresh cookies via browser automation. That adds latency from starting the browser + navigating just to fetch them.

Yes true. Imagine a checkout process. We will still cut 20 steps of filling the form

Yes agree on the api discovery part.

Amazing! So this will work 100% accurately every time?

Which website do you want to automate?

As a test... AWS Cost explorer

So you want to get your balance ?

Lets just say a monthly total for multiple accounts after all calculations are processed by AWS.

So we could do 1 tool to get balance for 1 account and call this multiple times

100% agree! Initial SSO access is required, but once the AWS console is available I assume the tool can learn and build the API endpoint. From there it becomes a repeatable process across multiple accounts and I assume it could mostly run in parallel.

Awesome yes you can provide the session cookie when you call the tool. We will include that in browser use