Combining Research, Data Processing, and Browser Automation

In the previous two guides, we covered agentic search across webpages and Spaces and natural-language web extraction across multiple pages. Real browser work often combines both with live webpage actions. A single task can start with reading several sites, switch to clicking page filters when details are hidden behind buttons, run budget calculations on the collected rows, adapt when requirements change mid-run, and fill the winning option into a web form where you are already signed in.

This guide walks through an end-to-end scenario that connects all of those steps in one Starizon conversation:

Walkthrough Scenario: Collect candidate technical workshops from multiple event websites, filter them by date, location, hands-on lab availability, and team budget, adjust the criteria mid-run ("keep weekend workshops only"), and fill the selected workshop into a signed-in internal training request form while stopping before final submission for your review.

1. How One Conversation Coordinates Research, Data, and Browser Actions

In Agent S6 mode, Starizon coordinates reading, extraction, calculations, and connected tools, and delegates live webpage interactions to its browser agent (Agent S6). Rather than forcing every step through screen-clicking, the conversation uses the right capability for each stage of the task:

Why Use Agent S6 Instead of /automation?

Starizon includes an /automation command for standalone browser routines on a webpage. However, /automation runs browser actions on their own without attaching your Knowledge Base collections, Spaces, uploaded spreadsheets, or Skills. Because our workshop workflow combines open event tabs, a saved policy collection, an uploaded budget CSV, and a signed-in request form, select Agent S6 in the mode menu (Modes and Context).

2. Prepare Your Project Materials Before Starting the Run

Set up the four inputs for this illustrative scenario in Chrome:

1. Candidate Event Tabs (@Event Sites): Open the three event directory and organizer tabs you want to evaluate and place them in a Chrome tab group named Event Sites.
2. Saved Policy Collection (@Team Event Policy): A Knowledge Base collection containing your team's training rules (workshops must take place in November in Seattle, Austin, or live online, stay within an $850 per-person budget cap, and offer a hands-on lab track).
3. Partner Discount Sheet (team-budget.csv): Use the Upload menu in the chat input to attach a CSV listing negotiated partner discount amounts by workshop_code (any workshop not listed in the sheet receives a $0 discount):

workshop_code,partner_discount_usd
WK-204,50
WK-311,0
WK-108,0

4. Target Internal Form (@Training Request Form): Open the internal portal tab where you are already signed in and where the training request will be entered.

In the chat input Tools menu, enable Knowledge Base and Web (plus Apps if you want to log the final choice to Google Sheets or Notion afterward) so the conversation can read your policy collection and check external organizer FAQ pages if an open tab omits ticket details.

3. Coordinate Reading, Browser Actions, and Iterative Calculations

Start the workflow in Agent S6 mode with a prompt that defines the collection scope, budget calculation, CSV export, shortlist review, and form-filling goals:

Using @Event Sites, our @Team Event Policy collection, and the uploaded team-budget.csv:

1. Collect November technical workshops from the first 2 pages of each tab in @Event Sites. If a page hides session dates, hands-on lab details, or material fees inside collapsed detail cards or behind a "Next" button, open those controls to collect the details.
2. Join the collected workshops with team-budget.csv on workshop_code (using $0 discount when a code is not in the CSV), calculate the total cost per person (base_fee + lab_fee - partner_discount_usd), and filter for workshops that meet @Team Event Policy and stay within our $850 cap.
3. Export the evaluated workshops as a CSV file, show a comparison table with workshop_code, title, dates, location, hands_on_lab, base_fee, lab_fee, partner_discount_usd, total_cost, and source_url, and pause so I can pick which workshop to submit.
4. After I choose a workshop, fill its details into @Training Request Form, and stop before submitting the form so I can review the completed fields.

How Starizon Loops Between Reading, Calculation, and Browser Actions

Rather than executing a rigid one-way pipeline, Starizon moves back and forth between capabilities as the task requires:

• Static reading first: On a static schedule page (Tab 1), workshop dates, cities, lab tracks, and fees are already visible, so Starizon reads the page content directly without clicking page controls.
• Interactive pagination when needed: On an interactive directory (Tab 2), Agent S6 clicks Next to collect the second page of listings.
• Looping back to the page when calculations spot a missing fee: Suppose the initial pass over WK-204 captures the $780 base ticket and notes "Hands-on lab available", but the lab fee column is blank. Instead of guessing $0, Starizon returns to the WK-204 card in the browser, expands its fee breakdown drawer to pull the $60 lab materials fee, and re-runs the budget calculation ($780 + $60 - $50 = $790).
• Filling gaps from linked pages: When Tab 3 omits whether a hands-on lab track is available, Starizon uses Web search to check the organizer's linked syllabus page before finalizing the table.

4. Adjust Requirements Mid-Run and Respond to Prompts

Suppose the workflow is running when your manager messages you with a new constraint: team members cannot miss weekday sprint reviews in November, so only weekend workshops (Saturday and Sunday) are allowed.

You do not need to discard your setup and start a brand-new conversation.

Send a Mid-Run Update

While the run is active, type the updated rule into the chat input and send it:

Update the filter: only keep workshops held on weekends (Saturday or Sunday). Exclude any session that requires Friday attendance.

Check that your update was delivered in the conversation thread, and verify in the revised shortlist that workshops requiring Friday attendance were excluded.

Three Ways to Interact With an Active or Paused Run

During a multi-step workflow, you have three distinct controls depending on what is happening on screen (Agent S6):

5. Verify the Extracted Shortlist and Select the Winner

Here is an illustrative example of the shortlist table and CSV output produced after applying the weekend-only rule and joining with team-budget.csv (Total Cost = Base Fee + Lab Fee - Discount):

Before selecting WK-204 in the question card, run a quick check on the intermediate output:

1. Check the conversation report and exported CSV: Confirm in the conversation summary that all three tabs in @Event Sites were read without errors, and download the generated .csv file card to check the full row set (Extracting and Processing Data).
2. Spot-check the winning source page: Open the Source URL for WK-204 to confirm the Nov 14-15 weekend dates and the $60 lab fee on the organizer's page.
3. Select your choice: Choose WK-204 using the question card's control so Agent S6 proceeds to fill @Training Request Form.

6. Fill the Signed-In Form and Review Before Submission

In the final stage, Agent S6 switches to your open @Training Request Form tab (where you are already signed into your workplace portal) and enters the verified details:

• Event Name: Production Kubernetes & Observability Lab (WK-204)
• Dates: 2026-11-14 to 2026-11-15
• Location: Seattle, WA
• Requested Budget: $790.00 USD ($780 base + $60 lab fee - $50 partner discount)
• Justification: Matches @Team Event Policy (November weekend session, Seattle location, includes hands-on lab track).

Review the Completed Form and Confirmation Controls

When automating form filling, purchases, or record updates, build an explicit review stop into your request: ask Starizon to stop before submission so you can review the completed form. If a confirmation card appears, respond through its controls.

Keep two guidelines in mind during this step:

1. Use the confirmation controls when prompted: Inspect the recipient, field values, and selected options on the @Training Request Form tab first. When a confirmation card is waiting in the conversation, approve or cancel using that card's own controls rather than sending a separate plain-text message.
2. Do not type into the webpage form while the agent is operating it: If you want to change a field before submitting (for example, adding a cost-center code), ask Starizon to update that field or pause the run before typing into the form yourself.

7. Log to Connected Apps, Save to a Space, or Schedule Weekly Checks

Once the form is reviewed and submitted, you can extend the workflow across your connected tools and future sessions:

• Log the decision to Google Sheets or Notion: With Apps enabled in the chat input Tools menu (Apps and MCP), ask Starizon to append the selected WK-204 row and the evaluated table to your team's training tracker in Google Sheets or create a summary brief in Notion, then open the returned link to verify the entry.
• Save to a Space: Save the conversation, the exported CSV of evaluated workshops, the uploaded team-budget.csv, @Event Sites, and @Team Event Policy into a Space named Q4 Team Training (Spaces). Next month, select @Q4 Team Training in a new chat to filter the saved table under new constraints or re-read the live event pages for updated dates.
• Monitor Event Pages for Changes: If a preferred workshop is waitlisted or you want to watch an organizer page for newly added weekend dates, use Page Monitors (/monitor) to track page changes and get notified (Page Monitors).
• Schedule Recurring Checks: Use /schedule or Tools -> Schedule (Schedules) to run a recurring workflow, such as re-checking the event directory URLs every Monday morning against @Team Event Policy and team-budget.csv to surface newly added weekend workshops. Scheduled tasks reopen the saved page URLs and use the materials and tools configured on the schedule, running while Chrome is running and your device is awake (scheduled agent runs pause after 10 minutes, and paused runs expire after 24 hours).

Want to revisit a specific part of this workflow? Start from the beginning with Introducing Starizon: An Agentic Workspace for Chrome, review large-scale search across project materials in Agentic Search Across Multiple Webpages and Spaces, or dive into multi-page extraction in Natural-Language Web Extraction Across Multiple Pages.