Create Browser Automation via Dialogue
The Guide project demonstrates using dialogue to guide users in creating and adjusting browser workflows. Task: The Guide project uses a conversational agent to help users create browser automation routines. The workflow involves describing a task, recording a session in a cloud browser, and having the agent analyze the session to identify backend endpoints and extract parameters. The agent then builds a reusable routine, allowing users to refine it through chat and review diffs without manual code editing. Human involvement is required to define the initial task scope and approve the generated workflow steps.; tools: Tool described in the source. Human role: Not stated in the source; limitations: This is a self-published demo on Hacker News and cannot be treated as a verified production case study. The approach relies on reverse-engineering websites into reusable APIs, which may fail on sites with complex or non-standard backend structures. Integration requires access to the Vectorly cloud browser and the Guide agent. Users must manually review generated diffs to ensure security and correctness before deploying routines..
Tools
Tool described in the source
Published on Agent Work Atlas:
Overview
The Guide project demonstrates using dialogue to guide users in creating and adjusting browser workflows.
Workflow
The Guide project uses a conversational agent to help users create browser automation routines. The workflow involves describing a task, recording a session in a cloud browser, and having the agent analyze the session to identify backend endpoints and extract parameters. The agent then builds a reusable routine, allowing users to refine it through chat and review diffs without manual code editing. Human involvement is required to define the initial task scope and approve the generated workflow steps.
Reported result
The source describes this as a demonstration of a new feature for the Vectorly open-source app. No independent verification of production deployment, specific efficiency metrics, or error rates is provided in the text. The actual outcome is limited to the author's claim that the agent can build routines from recorded sessions. A measurable acceptance check would be to verify if a recorded session successfully generates a functional API routine for a specific website.
Human role
Not stated in the source
Limitations
This is a self-published demo on Hacker News and cannot be treated as a verified production case study. The approach relies on reverse-engineering websites into reusable APIs, which may fail on sites with complex or non-standard backend structures. Integration requires access to the Vectorly cloud browser and the Guide agent. Users must manually review generated diffs to ensure security and correctness before deploying routines.
Try it
I want to test the Guide agent from Vectorly to create a browser automation routine. First, I will describe a simple task: [replaceable task description, e.g., 'fetch the latest 5 headlines from a news site']. I will record my interaction with the target website [replaceable URL] in the Vectorly cloud browser. Then, I will ask the Guide agent to analyze the session and build a reusable routine. I will review the generated code and diffs to ensure it correctly identifies the backend endpoints. Finally, I will run the routine to verify it returns the expected data. Please confirm the steps before I proceed with the recording.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.