Browser Automation Helper

Repetitive browser tasks — submitting expense reports, downloading monthly statements, filling intake forms with the same data, taking weekly screenshots of dashboards — eat hours over a year. The browser extension lets Flowly drive Chrome: open tabs, click, fill, extract. You describe the task, the agent does it.

What it does

• Open URLs in your existing browser session (not a headless one — uses your logins)
• Click, type, scroll, screenshot
• Extract structured data from rendered pages (tables, lists)
• Save data to memory or files
• Replay sequences as named workflows ("monthly expense report")

What you'll need

• Flowly browser extension (Chrome / Brave / Edge — install from the dashboard)
• browser_tab tool — built into Flowly
• A clear, repeatable task to start with — start small
• Your existing browser session signed into the relevant services

Setup

1. Install the extension

In the dashboard, Browser Extension tab → install. The extension authenticates against your Flowly server; once paired, the browser_tab tool can drive it.

2. Verify the connection

Agent calls browser_tab(action="list") and replies with your tabs. If this works, the extension is wired.

3. Try a simple sequence

"Take a screenshot of my dashboard at app.useflowlyapp.com/dashboard and save it to ~/screenshots/dash-<today>.png"

Agent:

1. Opens the URL in a new tab (browser_tab(action="open", url=...))
2. Waits for page load
3. Takes screenshot (browser_tab(action="screenshot"))
4. Saves the image to disk (write_file with base64 decode)

4. Build a named workflow

For repeated tasks, define a script the agent saves:

Now whenever you say:

"Run monthly-expense-report — description: Brno conference travel, amount: $432, date: 2026-04-12, category: Travel"

The agent runs the saved workflow with your inputs.

5. Scrape data

"For each row in the table at internal.example.com/users, extract user_id, name, email. Save to ~/users.csv"

Agent:

1. Opens the URL
2. Waits for table render
3. Extracts via DOM selectors (or asks for the selector if ambiguous)
4. Writes CSV

Pitfalls

• Authentication. The agent uses your browser session. If you're logged out, it can't drive. Re-auth and retry.
• Captchas / 2FA. Anything human-attestation bound is impossible for the agent. It will pause and tell you.
• DOM changes break workflows. When the target site updates its HTML, your saved workflow breaks. The agent should detect "selector not found" and ask you to re-record vs guess.
• Don't automate destructive actions blindly. "Cancel my subscription on every site" is the kind of automation that ends badly. Confirm-and-show before submit on anything destructive.
• Rate limiting. Don't hammer a site. Add explicit pauses between operations. Your IP can get blocked otherwise.

Tips

• Start tiny. First workflow: just open a URL and screenshot. Build up to multi-step from there.
• Test on staging. If the target service has a staging URL, test there first. Don't blow up production data while debugging.
• Save workflows by name. "Monthly expense" is more useful than rewriting the whole thing each time.
• Pair with cron. "Take a screenshot of my dashboard every morning at 9" → cron + browser_tab → daily PNG in your folder.
• The agent isn't a hacker. Don't ask it to bypass logins, scrape rate-limited data, or do anything you couldn't do yourself in the browser.
• Privacy: be aware. The agent can see whatever your browser can. Logged-in banking sites, private DMs, anything. Don't share workflow definitions with other people.