I'm an engineer. I love solving problems. When working on an AI agent I discovered that LLM are really bad at web automation. They know what steps to perform to complete a task, they just don't know how to do them. I saw this as a fundamental problem that had to be solved before I could create a general purpose agent. So I solved it.
The solution is webSlinger. This is how it works. You show webSlinger how to perform the task in your browser and webSlinger copies you in real time in another tab. You get immediate feedback that webSlinger can perform each action as you do it. Save the resulting "session map", and now webslinger can translate from "what to do" to "how to do it." The LLM then has what it needs to generate an automation script.
Plan -> Record -> Generate -> Automate (repeatedly).
I built in the capabilities I cared about the most, secure login and MFA with locally held credentials and array extraction.
Websites often display lists, tables or repetitive structures like job search results or real-estate listings. One thing that webSlinger excels at is extracting such data in just a few clicks. Right-click on an instance of the array, check the boxes corresponding to the data you want, and click Record. The array is extracted with field names matched to your goals or semantically discerned.
What array-like data would you like to extract from a site?
